main
641 Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
 mchaitra
|
fa72a1ed57 | Add 04-migrate-to-new-namespace directory for registry namespace migration sample (#1886) | ||
|
Guruprasad
|
9018c5e3c7 |
fix: self-managed strategy sample - remove invalid namespaces, fix IA… (#1858)
* fix: self-managed strategy sample - remove invalid namespaces, fix IAM and CLI docs (#1856) * fix: rename namespaces to namespaceTemplates per reviewer feedback |
||
|
Rui Cardoso
|
ee31c1790e |
feat(harness/14-s3-filesystem): optional script to provision the S3 Files prerequisites (#1881)
Both scripts in this sample mount an existing S3 Files access point, and nothing in the folder creates one — so on a fresh account --access-point-arn, --subnet-ids and --security-group-ids have nothing to point at and the sample cannot be run at all. Adds an optional provision_s3_filesystem.py that creates the prerequisites (versioned bucket, service role, file system, access point, a mount target per subnet, and an NFS security group) and prints the exact command line to paste into either script. The sample scripts themselves are unchanged. Networking is bring-your-own by default: the script discovers private subnets that already have NAT-gateway egress rather than creating a VPC, because a NAT gateway bills hourly. --create-vpc is opt-in. Every resource is recorded in provision_state.json as it is created, and --teardown deletes only what is recorded there. |
||
|
Rui Cardoso
|
f67f2c041f |
fix(harness/14-s3-filesystem): missing IAM permission, leaked execution role, and false-pass command guards (#1880)
The sample could not create a harness at all: the execution-role policy was missing s3files:ListMountTargets, which the runtime validates at create time, so CreateHarness was accepted and then landed in CREATE_FAILED. Also fixes an execution-role leak (the shared HarnessExecutionRole and its inline policies survived every run, including successful ones, breaking the next sample in the folder), several guards that reported success on a failed or never-executed VM command, and a mid-stream read timeout that discarded an answer already in hand. README: document the bucket-versioning requirement, the real IAM scope, the 2-3 minute VPC-mode create time, and the flags that were undocumented. |
||
|
Rui Cardoso
|
2ac3ce2b35 |
fix(harness/13-aws-skills): handle transport and throttling stream errors (#1879)
`stream_response` caught only `EventStreamError`, so a mid-stream transport failure (`ReadTimeoutError`, `ConnectionClosedError` — both `BotoCoreError`) or a `ThrottlingException` (a plain `ClientError`) escaped as an unhandled traceback that named nothing about the cause, aborting the demo even when the agent's complete answer had already been printed. Catch both base classes, report the cause, and re-raise only when no content arrived. Also fix `--raw-events` discarding good responses: raw mode skipped text accumulation, so `full_text` stayed empty and the "did we receive anything?" guard re-raised even a cosmetic close error. Remove the dead `REGION` assignment (never read; the region actually resolves in `utils/client.py`, which honours `AWS_DEFAULT_REGION` or `AWS_REGION`) along with the now-unused `import os`, and the entirely unused `import boto3`. Correct a comment that described a `tools`/`allowedTools` argument the call does not pass. README: add a Prerequisites section including the boto3 >= 1.43.32 floor that `awsSkills` requires, document `--role-arn`/`--raw-events`/`--skip-cleanup` and the ~2-3 minute wait for READY, and state the tagged-union and wildcard semantics of `skills` entries. |
||
|
Rui Cardoso
|
34e039f36a |
fix(06-async-step-function): report failures instead of SUCCEEDED, and fix the sample test path (#1878)
The Step Functions workflow routed a caught harness error to a Pass/End state, so an execution whose InvokeHarness failed with AccessDeniedException finished SUCCEEDED with nothing written to DynamoDB. The missing memory/event permissions that caused that failure are also fixed, along with the JSON extraction (which crashed on bare JSON and silently truncated nested objects), the retry policy (which billed caller-fault errors four times), the --use-samples loop (which aborted on the first sample because of word-splitting), both poll loops (which treated ABORTED/TIMED_OUT as success), and three unreachable error handlers. All fixes reproduced against upstream main and re-verified live in a real AWS account over 3 deploy/teardown cycles. |
||
|
Rui Cardoso
|
a9b2c0c920 |
fix(01-harness): stop 07-oauth reporting success for a failed invoke, and leaking on failure (#1877)
InvokeHarness over HTTPS returns HTTP 200 and reports agent-side failures as
{"message": ...} frames inside the event stream. The parser read only
obj["delta"], discarded those frames, printed "(No text deltas found in
stream)", and then printed the full success narrative claiming all three auth
hops had worked. Separately, the script had no try/finally, so a failure
anywhere between Step 1a and Step 4 left two Cognito pools, a credential
provider, a Lambda, a gateway + target, a harness and three IAM roles alive.
Correctness:
- Raise on non-200, collect and raise on stream error frames, and raise when
there are neither deltas nor an error frame.
- Grant the 8 memory/event actions the harness needs at runtime; CreateMemory
and ListMemories take no memory ID, so they must be scoped to "*".
- Replace three fall-through polls with deadline loops that raise on terminal
status and on timeout.
- Wrap provisioning in try/finally so cleanup always runs.
- _wait_gone no longer reports success on timeout.
- Do not print "Cleanup complete!" when resources were skipped.
- _ensure_policy publishes a new default version instead of swallowing
EntityAlreadyExists, pruning the oldest non-default at the 5-version limit.
- Paginate the Cognito pool lookup and every AgentCore list call
(ListHarnesses, ListGateways, ListGatewayTargets) through one helper.
- Match the gateway and target conflict fallbacks on name; the target fallback
took the first target it saw, which returns the wrong id on a gateway that
carries any other target.
- Drop the N+1 get_gateway in the gateway fallback; ListGateways returns name.
- _del_role can now delete a policy that has non-default versions.
- Convert six `except Exception: pass` handlers into reported skips.
Docs and lint:
- README: document the IAM permissions the walkthrough needs, and drop the
`pip install requests` step -- requests is already in ../../requirements.txt.
- utils/lambda_function_code.py: fix a latent import-order violation and a
docstring that named the wrong tools.
|
||
|
Rui Cardoso
|
9b6ec3f736 |
fix(05-agent-skills): stop reporting success the sample never checked (#1876)
The demo prints four green ticks after commands whose exit codes are never read, and states a sheet count for a file nothing ever opened. - Replace two hand-rolled `for i in range(12)` polls with the shared `poll_harness_status`. Both had no `else` and no `raise`, so on timeout they fell through silently and `update_harness` ran against a CREATING harness (`ConflictException`). Real create time is ~95-133s; the ceiling was 60s. - Wrap the lifecycle in `try/finally`. The file had zero `Try` nodes, so the crash above leaked a billable harness and the shared execution role. - Raise on a failed skill install instead of printing the tick regardless, and gate on the verification command's answer instead of discarding it. - Add `run_command`, returning stdout, stderr and `contentStop.exitCode`, so a missing file is distinguishable from an unreadable one. - Validate downloads: `.xlsx` is a zip, so `zipfile.is_zipfile` rejects non-workbooks and the sheet count is read from the file rather than hardcoded. - README: document the three command outputs, the LibreOffice install the agent performs on `node:slim`, and the new diagnostics. |
||
|
Rui Cardoso
|
d35540c0c5 |
fix(01-harness): stop 04-mcp-integration reporting success for a report it never retrieved (#1874)
Part 5 read the research report with `cat ... 2>/dev/null || echo '{}'`. When the
agent answered in prose without writing the file, cat's error was discarded and
the fallback emitted `{}` — valid JSON — so json.loads succeeded, the
JSONDecodeError branch never ran, and the sample printed "Research Report
Generated:" followed by an empty object. Ask for the file plainly and keep both
halves of the answer: stderr says why it is missing, the exit code says whether
to trust stdout. A truncated file now reads differently from an absent one.
stream_invoke iterated the event stream bare. internalServerException,
validationException and runtimeClientError are modelled `exception: true`, so
botocore sets a 400 and raises EventStreamError out of the iterator rather than
yielding a dict — the `elif "internalServerException" in event` branch could
never run, and any stream failure escaped into `finally`, ending the demo at
Part 1. Catch (ClientError, BotoCoreError): both are needed and neither covers
the other, since EventStreamError subclasses ClientError while ReadTimeoutError
subclasses BotoCoreError.
That makes Part 4's `Errors encountered:` reachable for the first time. It used
to sit above the except clause every real run took, and could only have printed
0 anyway, because the sole thing appending to that list was the unreachable
branch above. It now prints the service's own message, with an explicit else so
a bad URL that starts succeeding shows up instead of passing silently.
Part 3 is titled "MCP with Authentication" and printed "configured with the API
key", but the only line that would use the key was commented out — so it
demonstrated a config with no credential in it. `headers` is a real member of
remoteMcp (map<string,string>). Enabling it means the config dump below carries
a secret, so print a deep copy with every header value masked, honouring the
comment already there forbidding echoing the key. Masking by iteration, not by
header name, so switching to x-api-key needs no change here.
README: the `curl` used to test an MCP server failed against a healthy one —
missing the `Accept` header (406) and not a complete JSON-RPC request (-32700).
It appeared twice, once as the first troubleshooting step. The
`internalServerException in the stream` entry described an event that never
arrives; replaced with one keyed on the EventStreamError that does. Documented
`headers` properly, and added an entry for Part 5's new missing-report message
so it is not mistaken for a bug.
|
||
|
Rui Cardoso
|
ceacf51e09 |
fix(01-harness): correct the gateway docs, harden the pollers, stop leaking the role (#1870)
The gateway-integration sample runs correctly end to end, but its docs describe a different sample than the one that ships, its gateway pollers cannot report the failures the service actually returns, and cleanup leaves the IAM role behind. Documentation - The README walked the reader through a "Create routing rule" step. A gateway routes to its targets as soon as they report READY -- no rule is needed to run this sample. Replaced with a note explaining that rules are a separate, optional feature for shaping traffic across multiple targets. - The Key Concepts heading read "IAM auth (`NONE` type)", which contradicts itself: NONE means no inbound authorizer, and AWS_IAM is a distinct value the sample does not use. Replaced with a "Two independent authorization sides" section and a table separating inbound (authorizerType on create_gateway) from outbound (credentialProviderConfigurations on create_gateway_target), so it no longer conflicts with 07-oauth, which configures both. - The README flow and the module docstring were numbered 1-7 while the script prints Step 0-4 plus cleanup. Renumbered both to match the real output. - Both described Step 3 as creating a harness "wired to the Gateway". No such binding exists: create_harness takes no gateway parameter, and the gateway ARN travels per-invoke in the `tools` argument to invoke_harness. Corrected, and stated explicitly, because the false coupling is what made the Clean Up section's ordering advice look justified. - The Clean Up snippet listed the deletes in an order that contradicted its own "targets before gateway" note, omitted the role deletion entirely, and implied the harness must go first because it references the gateway. Only the target-before-gateway constraint is real. - Added troubleshooting entries for HTTP 429 (the default Exa endpoint is keyless on a shared, rate-limited free tier, so a busy account can exhaust it; shows how to attach your own key as an outbound API_KEY credential provider), for the UPDATE_UNSUCCESSFUL / SYNCHRONIZE_UNSUCCESSFUL statuses, and for a target stuck in CREATE_PENDING_AUTH. Code - GATEWAY_POLL_TIMEOUT was 120s while the README told the reader to expect "2-3 minutes" -- the same contradiction between a local constant and the documented wait that #1863 fixed for the harness poller. Raised to 300s, matching 07-oauth's budget for these two operations. Measured: both the gateway and the target reach READY in about 5s, so the old ceiling only ever mattered on the slow runs it was least able to survive. - The pollers checked for the wrong terminal statuses. A gateway reports UPDATE_UNSUCCESSFUL, not FAILED, when an update fails, and a target adds SYNCHRONIZE_UNSUCCESSFUL; neither was treated as terminal, so a failed resource was polled until the timeout and the service's statusReasons was never shown. The target poller also tested for DELETE_FAILED, which is not in the target status enum at all -- a dead branch. Both now use the real enums, and the three *_PENDING_AUTH states fail fast with an explanation instead of spinning on a state that cannot clear by itself. - The stream handler surfaced only internalServerException. validationException and runtimeClientError are modelled stream events too, and falling through them printed nothing, so a rejected invoke looked like an agent with no answer. - REGION was read from AWS_DEFAULT_REGION alone, while utils/client.py deliberately honours AWS_REGION as well and treats an empty string as unset. A shell exporting AWS_DEFAULT_REGION="" therefore built a working harness_control and a gw_control that died on "Invalid endpoint: https://bedrock-agentcore-control..amazonaws.com". Now imports the shared REGION so both clients resolve identically. - _cleanup now deletes the execution role, guarded by created_role so that a role supplied via --role-arn is never touched. The role name is shared by every sample in this folder, so deleting a caller's role would destroy something the script does not own. Without this the role outlived every run. - Annotated the create_gateway and create_gateway_target call sites with which authorization side each parameter controls, dropped the now-unused os import, and moved the utils.iam import back into alphabetical order. Every API-shape claim added here was read off the botocore service model rather than inferred: the authorizerType and credentialProviderType enums, the apiKeyCredentialProvider members, the credentialLocation enum, the gateway and target status enums, the InvokeHarness stream event members, and the absence of any gateway parameter on CreateHarness. Verified live in us-west-2, five end-to-end runs, the last on the exact bytes committed here: gateway and target reach READY with no routing rule, the agent calls exa-search___web_search_exa through the gateway and returns real results, and teardown removes the harness, target, gateway, inline policy and role, leaving the account clean. Unit-checked against stubbed clients: both pollers for the happy, every terminal, every pending-auth, transient-passthrough and timeout paths; the stream handler for all three error events, tool-use, raw mode and the tools payload shape; and _cleanup for created_role=True, created_role=False, the default argument, and failure before any resource exists. The empty-region regression was reproduced before the fix and confirmed resolved after. Ruff check and format both pass; findings on the file go from 3 to 0. |
||
|
Rui Cardoso
|
cce2ac62a6 |
fix(01-harness): stop 01-custom-containers prompting for tools its images lack (#1868)
* fix(01-harness): stop 01-custom-containers prompting for tools its images lack
The custom-container sample asks the agent to run three tools that are not
present in the images it attaches, and documents the resulting output as the
expected behaviour. Live runs of all three presets:
- python preset told the agent to `curl localhost:3000`, but
python:3.12-slim ships neither curl nor wget. Now asks for urllib.
- go Part 5 asked for the architecture "using 'file' command", but
golang:1.24 has no `file`. The agent burned a turn on the failure and
recovered with readelf ("Let me try with readelf as an alternative"),
so the demo passed while doing something other than what it documents.
Now asks for `ls -lh` + `readelf -h` directly: 3 shell calls -> 2.
- README's Architecture block advertised "use curl" for every image and
described the microVM as Node-only. It also documented no python prompt
at all, though --language python is a supported preset.
Also retries the first invoke_harness on RuntimeClientError (3 attempts,
15s apart). A harness reporting READY does not guarantee the container's
microVM answers health checks yet: one python run died with "Runtime
health check failed or timed out" while that runtime's own CloudWatch log
group recorded a clean startup and zero errors. Only the call is retried,
never a stream that already emitted output, so no text can print twice.
Verified live in us-west-2, all three presets deployed, invoked and torn
down (exit 0):
- go readelf on the first try, no fallback turn; ELF64 x86-64
- python agent used urllib, HTTP 200, no curl attempt
- node unchanged default path, node v26.5.1 / npm 11.17.0, chalk ok
Harness naming follows --language so Go/Python runs no longer create
resources labelled NodeContainer_*, which made leaked resources and their
auto-named managed memory indistinguishable in the console.
* style(01-harness): satisfy ruff format on the retry message
The `ruff format --check` step of the python-lint workflow rejected the
multi-line print() added for the RuntimeClientError retry: it used hanging
indent continuation where the formatter wants the parenthesized form.
Formatting only — the AST is identical before and after, so behaviour is
unchanged and the rendered message is byte-for-byte the same.
|
||
|
Bharathi Srinivasan
|
ec078649fa |
fix(os-actions): remove trailing slash from base_url to prevent doubl… (#1867)
* fix(os-actions): remove trailing slash from base_url to prevent double-slash 404s * fix(os-actions): suppress ruff BLE001/S110 on best-effort cleanup handlers |
||
|
Bryan Conklin
|
1226ec55aa |
fix(01-harness): inherit the shared harness poller in gateway-integration (#1863)
* fix(01-harness): inherit the shared harness poller in gateway-integration The gateway-integration sample fails on a first run and leaves the harness behind when it does. Two defects, both consequences of this file carrying its own copy of poll_harness_status instead of using the shared helper introduced in #1857: - The local HARNESS_POLL_TIMEOUT was 120s against provisioning measured at ~150s on the public network and ~255s in VPC mode, so the sample raises TimeoutError before the harness has had time to come up. - The local poller treated "FAILED" and "DELETE_FAILED" as its failure states. The status enum is CREATING, CREATE_FAILED, UPDATING, UPDATE_FAILED, READY, DELETING, DELETE_FAILED — there is no plain FAILED, so a harness that failed to create was polled until the timeout and the service's failureReason was never surfaced. Deleting the local copy and importing utils.harness.poll_harness_status fixes both, and stops the constant drifting from the shared value again. Cleanup had a third, independent defect. A harness cannot be deleted while it is still CREATING, and the finally block runs at exactly the moment that is most likely to be true — after a step failed mid-provisioning. The single delete_harness call was rejected with ConflictException, swallowed by a broad except, and the harness was left behind counting against the account's harness quota with nothing pointing at why. _delete_harness now waits out ConflictException on the same 600s budget, returns immediately on any other error code, and no longer catches BaseException-wide. README: the troubleshooting entry told readers to raise a constant this file no longer defines. Replaced with the real timeout and the CREATE_FAILED / failureReason check, plus a note that the retry-delete log line is expected. Verified: module imports and resolves poll_harness_status from utils.harness at 600s; _delete_harness checked against stubbed clients for the success, retry-then-succeed, non-conflict, timeout, and already-deleted paths. Ruff findings on the file go from 4 to 3 — the remaining three are pre-existing and untouched, per the guidance not to reformat unrelated code. Related to #1857, which fixed the same class of defect in 00-getting-started and noted that the other samples in this folder share it. * chore: add Bryan Conklin to CONTRIBUTORS.md |
||
|
Rui Cardoso
|
065c742811 |
fix(01-harness): fix memory lifecycle, share the poller, and stop leaking the role in 3 use-case samples (#1866)
The three use-case samples under 02-use-cases had the same family of provisioning-lifecycle bugs the advanced examples had (#1864), plus a memory-specific cluster in the travel agent. aws-builder-agent: - Never deleted the shared execution role at all — it imported only create_harness_role. Because every sample in this folder shares one role name (HarnessExecutionRole), the leftover role poisons the next sample's run. Now imports delete_harness_role and deletes it in finally, guarded by a created_role flag so a caller-supplied --role-arn is never destroyed. - Dropped its private poller for the shared utils.harness.poll_harness_status. The local copy checked for a "FAILED" status that is not in the enum (CREATE_FAILED / UPDATE_FAILED / DELETE_FAILED), so it could never detect a real failure, and its 180s ceiling left little margin over the ~145s a harness actually takes to reach READY. webapp-visual-testing: - Replaced both fixed 24×5s (120s) poll loops with the shared poller. The old ceiling expired while the harness was still CREATING — the live run needed ~29 polls (~145s) to reach READY — after which the update_harness call that attaches the Node container fails with ConflictException. travel-agent (memory cluster): - create_memory / get_memory return the resource under a top-level "memory" key only, so the old `resp.get("id")` first branch was dead code — proven live (it returns None). Now reads resp["memory"]["id"] / ["arn"]. - The memory status enum is CREATING → ACTIVE with a plain FAILED and no READY (unlike the harness). The old loop waited for ("ACTIVE","READY") and never checked FAILED, so a failed memory was polled to the ceiling and its failureReason never surfaced. New poll_memory_status honors the real enum and raises on FAILED. - Removed the adopt-then-delete fallback that matched any memory whose id contained "TravelGuide" and deleted it in cleanup — that destroyed a memory the sample did not own. memory_id is now only set when this run created the memory; a pre-existing one raises ConflictException, which is caught and the memory demo is skipped with a clear message. - Replaced the harness create-wait (a 12×5s = 60s ceiling, versus the ~145s observed) and the post-update wait with the shared poller. - Cleanup now deletes the harness before the memory it references, each guard independent so one failure cannot skip the role deletion. - README updated to match the new ConflictException behavior. Also cleared pre-existing lint debt on the touched files (import order, dict()→literal, blind-except noqa) and removed a stray shebang (EXE001). Tested live in a fresh account (us-west-2): baseline runs reproduced the role leak and the dead memory-id branch; all three fixed samples ran end to end (builder scaffolds a project, travel completes all 6 parts incl. multi-turn memory recall, webapp captures 3 Puppeteer screenshots) and left zero leaked resources — role, harness, and sample-created memory all deleted. |
||
|
Rui Cardoso
|
d74b972283 |
fix(01-harness): correct execution limits, share the poller, and stop leaking the role in 4 advanced samples (#1864)
* fix(01-harness): correct execution limits, share the poller, and stop leaking the role in 4 advanced samples
Follow-up to #1857, applying the shared poll_harness_status helper and the
role-lifecycle fixes to four advanced samples that need no extra infrastructure,
and fixing a factual error in the execution-limits demo.
- 03-execution-limits: the demo claimed maxTokens truncates the response
(maxTokens=10 -> "~10 tokens"). On InvokeHarness maxTokens is a *per-iteration*
ceiling, not a budget for the whole invocation, so the agent keeps going across
iterations and the answer is not truncated. Verified live: maxTokens=256 still
produced 1413 output tokens and stopped with end_turn. Corrected the code, the
docstring and the README, and switched the demo to compare per-run output
tokens. maxIterations and timeoutSeconds do stop the agent
(max_iterations_exceeded / timeout_exceeded), which the demo still shows.
- All four (01-custom-containers, 03-execution-limits, 04-mcp-integration,
13-aws-skills): replace the local wait-for-READY loop with the shared
poll_harness_status from #1857. The local loops had a fixed ~60s ceiling on an
operation that takes ~150s (and longer for a container pull or AWS-skills
harness), and several checked for a "FAILED" status that is not in the enum
(it is CREATE_FAILED / UPDATE_FAILED / DELETE_FAILED), so they could never
detect a failure.
- 03-execution-limits, 04-mcp-integration: role and harness creation were above
the try block, so a CreateHarness failure exited leaving the shared IAM role
behind. Moved creation inside try/finally and made cleanup delete each resource
independently, matching #1857's getting-started fix. (01-custom-containers and
13-aws-skills already create inside try and gain a created_role guard so a
caller-supplied --role-arn is never deleted.)
Tested live in us-west-2 on an account with no pre-existing Harness resources:
each sample provisioned, ran its demo end to end (custom container attach +
npm install; MCP tool load and the invalid-URL error path; AWS Skills glob
load), and cleaned up with no leaked role or harness.
* fix(01-harness): stop logging the MCP API key in mcp-integration
Part 3 printed the first 8 characters of MCP_API_KEY. Eight characters of a
bearer credential is a meaningful leak, and sample output routinely lands in
terminal scrollback, CI logs and screenshots.
The line carried a `# codeql[py/clear-text-logging-sensitive-data]` comment,
which is legacy LGTM syntax that GitHub code scanning does not honour, so the
alert was live rather than suppressed. Print the key length instead of any part
of the value, and drop the comment that was not doing anything.
CodeQL flagged this as new on #1864 because wrapping the module in try/finally
re-indented the line; the expression itself is unchanged since
|
||
|
Visakh Madathil
|
4804e39267 |
fix(observe-evaluate): unblock span redaction and batch evaluation (#1861)
* fix(observe-evaluate): unblock span redaction and batch evaluation Two independent runtime failures in the observe/evaluate samples: 1. 01-observe/attribute_redaction.py — SensitiveDataRedactor.on_end assigned span._attributes[attr] directly, but span.attributes is a BoundedAttributes instance that is immutable once the span has ended, so redaction raised TypeError and the agent never completed. Rebuild a plain dict with sensitive values masked and reassign span._attributes. Reassigning avoids depending on the private inner storage name, which differs across OTel SDK versions (_dict vs _attributes). 2. 02-evaluate/utils/deploy.py — deploy never wrote otel_service_name to agent_config.json, so simulate.py omitted serviceNames from start_batch_evaluation, causing ParamValidationError. Write "<agentRuntimeName>.DEFAULT", matching the service name AgentCore emits on spans and the value sibling evaluate.py scripts derive from the ARN. Validated on live AgentCore: deploy writes the service name, spans land in aws/spans under that name, and start_batch_evaluation completes 5/5 sessions. * fix(observe-evaluate): resolve ruff lint errors in changed files - Sort import blocks (I001) - Annotate SENSITIVE_ATTRS as ClassVar (RUF012) - Drop unused E402 noqa directives; E402 is ignored repo-wide (RUF100) - Mark intentional broad-except catches with noqa: BLE001 (BLE001) --------- Co-authored-by: Visakh Madathil <visakhm@amazon.com> |
||
|
Rui Cardoso
|
e0ee66c644 |
fix(01-harness): make getting-started work on a fresh AWS account (#1857)
The getting-started sample fails on an account with no pre-existing Harness resources, and leaves the account in a state the other samples cannot recover from. The wait-for-READY loop was `for i in range(12)` with a 5s sleep -- a hard 60s ceiling on an operation measured at ~150s -- and it fell through instead of raising, so the failure surfaced one step later as `ValidationException: Harness is not READY`. The script then exited without deleting the IAM role it had created. Every sample in this folder shares the role name HarnessExecutionRole and create_harness_role() returned early when the role already existed, so that stale role went on to break the next sample too. Changes: - utils/harness.py (new): one correct poll_harness_status(). The same loop appears 16 times across this folder in 4 incompatible variants, most checking for a FAILED status that is not in the enum (CREATE_FAILED / UPDATE_FAILED / DELETE_FAILED), so they cannot detect failure at all. - 00-getting-started/getting_started.py: use the shared poller; move harness creation inside the try block so a CreateHarness failure cannot leak the execution role created just above it; clean up each resource independently. - utils/iam.py: least privilege (scoped Bedrock and CloudWatch Logs resources, 18 enumerated AgentCore actions in place of *Memory*-style wildcards that silently included Delete*, aws:SourceAccount on the trust policy) and full idempotency, so re-running repairs an existing role. Deletion now removes every attached policy, which a leftover would otherwise block with DeleteConflictException. - utils/client.py: honour AWS_REGION as well as AWS_DEFAULT_REGION, and default data-plane read_timeout to 900s -- botocore's 60s default is shorter than one agent turn and cut the response stream off mid-flight. - 00-getting-started/README.md: "Wait up to 60 seconds" contradicted the ~150s reality; also documents iam:UpdateAssumeRolePolicy. - requirements.txt: boto3>=1.43.32, the first release whose bedrock-agentcore model knows awsSkills. Tested live in us-west-2 on an account with no pre-existing Harness resources: full getting_started.py run passes all six steps and cleans up; a deliberately failed CreateHarness now deletes the role instead of leaking it; the scoped IAM policy brings a harness to READY, invokes it and writes logs under the scoped prefix; a degraded role is repaired by re-running. Plus 24 offline unit checks and a no-regression pass over the other 18 samples. Co-authored-by: shraone477 <308823281+shraone477@users.noreply.github.com> |
||
|
Diego Brasil
|
392e84238c |
fix(memory): unblock three first-run failures in AgentCore Memory samples (#1847)
* fix(memory): wait for IAM propagation before CreateMemory in boto3 quickstart
`00-getting-started/04-quickstart-boto3.py` creates the
`AgentCoreMemoryExecutionRole` and then calls `create_memory` a few lines later.
On a fresh account IAM has not propagated yet, so the service cannot read the new
role back and rejects the call:
botocore.errorfactory.ValidationException: An error occurred
(ValidationException) when calling the CreateMemory operation: Validation
failed during CreateMemory: Please provide a role with a valid trust policy
The message points at the trust policy, but the trust policy is correct. Verified
by reading the role straight out of IAM after the failure (principal
`bedrock-agentcore.amazonaws.com`, action `sts:AssumeRole`, unchanged) and then
retrying `create_memory` with the same ARN ~30s later, which succeeded.
This only affects the first run. Once the role exists, the
`get_role`/`NoSuchEntityException` guard skips creation and every later run
passes, so the bug self-heals and then stays hidden — which is why it reads as a
flake. New users on a clean account hit it every time, and the error sends them
off auditing a policy that was never wrong.
Sleep for 10s inside the `except NoSuchEntityException` branch so the delay is
paid only on the run that actually creates the role. Warm runs take the `get_role`
path and are unaffected.
Tested against us-east-1 by deleting the role first to reproduce cold-start
conditions:
- before: ValidationException on `create_memory`
- after: `Created: QuickstartMemory-SXxovF2FfB` / `Status: ACTIVE`
`ruff check` and `ruff format --check` both pass on the changed file.
* fix(memory): pin boto3/bedrock-agentcore floors for long-term memory samples
Three long-term memory samples call parameters and helpers that do not exist
in older boto3/bedrock-agentcore releases. Nothing in the tree declared a
floor, so an environment built a few weeks before a workshop fails on three
separate sub-features with errors that look like sample bugs:
02-long-term-memory/06-record-metadata/structured-metadata.py boto3
ParamValidationError: Unknown parameter in input: "indexedKeys"
02-long-term-memory/08-manage-extraction/skip-extraction.py boto3
ParamValidationError: Unknown parameter in input: "extractionMode"
02-long-term-memory/01-built-in-strategies/semantic.py sdk
TypeError: search_long_term_memories() got an unexpected keyword
argument 'namespace'
structured-metadata.py fails at 0 seconds on indexedKeys, which is item 3 of
its own "What you learn" list. skip-extraction.py fails after CreateMemory
succeeds, so it also leaks a billable memory resource.
Floors were established by bisecting the published wheels rather than
assuming, since the two botocore parameters land 34 patch releases apart:
botocore 1.43.0 -> CreateMemory has no indexedKeys
botocore 1.43.1 -> indexedKeys appears
botocore 1.43.34 -> CreateEvent has no extractionMode
botocore 1.43.35 -> extractionMode appears
bedrock-agentcore 1.9.0 -> search_long_term_memories() namespace_prefix only
bedrock-agentcore 1.10.0 -> namespace= appears (memory/session.py)
bedrock-agentcore 1.11.0 -> MemoryMetadataFilter has no build_expression
bedrock-agentcore 1.12.0 -> build_expression appears
boto3>=1.43.35 is the binding constraint because it requires
botocore>=1.43.35, which covers indexedKeys (1.43.1) as well. This agrees
with the existing 06-production-patterns/00-multi-region-replication pin of
boto3>=1.43.36, which also uses extractionMode.
bedrock-agentcore>=1.14 keeps the floor already documented in
01-built-in-strategies/semantic.py and 04-namespaces/README.md rather than
lowering it to the measured 1.12 minimum.
Verified against a live account (us-east-1) in a venv pinned to the exact
floors, boto3==1.43.35 / botocore==1.43.35 / bedrock-agentcore==1.14.0. All
three previously failing runs now exit 0:
structured-metadata.py boto3 -> EU metadata filter returned the expected
record with region/tier metadata
skip-extraction.py boto3 -> 12 events stored, 7 extracted, 4 skipped
semantic.py sdk -> all three semantic queries returned scored
records
Adds requirements.txt at the 02-long-term-memory level (the three affected
sub-features all live under it) and points the three affected README run
blocks at it. Scope is limited to what was executed and verified.
* fix(memory): replace retired Claude 3.5 Sonnet v2 model id in strategy overrides
02-strategy-overrides defaulted its override model to
anthropic.claude-3-5-sonnet-20241022-v2:0, which is retired. The sample is
the only place in the memory tree that lets you choose the extraction and
consolidation model, so the retired id makes the lesson unreachable: memory
creation succeeds, then extraction fails server-side and no records appear.
Confirmed retired against a live account (us-east-1):
anthropic.claude-3-5-sonnet-20241022-v2:0
-> ResourceNotFoundException: This model version is not supported
global.anthropic.claude-opus-4-6-v1
-> OK
MODEL_ID is a module-level constant read by _override_strategy(), which both
the boto3 and sdk surfaces call, so the single default covers both. The
OVERRIDE_MODEL_ID environment override is unchanged.
Verified end to end on the boto3 surface at the pinned dependency floors. The
override behaved exactly as the sample teaches, extracting only health facts
and suppressing the deliberate non-medical line:
[boto3] Medical facts (4):
- The user's mother had breast cancer at age 52.
- The user has type 2 diabetes.
- The user takes metformin twice daily for type 2 diabetes.
- The user is allergic to penicillin.
The Godfather mention correctly did not appear.
Also updates the two documentation occurrences that a participant would copy
and hit the same failure with: the AWS CLI walkthrough in this folder's
README (extraction and consolidation modelId) and the "Using Different
Models" snippet in the episodic healthcare example README.
Scope is limited to 01-features/04-manage-context-of-your-agent, the area
that was executed and verified. The same retired id still appears elsewhere
in the repo (05-authenticate-and-authorize, 06-workshops, an IaC doc table,
and two bedrock-models.yml files); those are intentionally left alone here
since they were not run.
|
||
|
Deepaxs
|
cd78928ae1 | Add payment skill orchestration docs for coding agents (#1815) | ||
|
satveerkhurpa
|
568ee829fe |
Add certificate-based auth (PRIVATE_KEY_JWT) sample for Okta and Entra ID (#1839)
* Add certificate-based auth (PRIVATE_KEY_JWT) sample for Okta and Entra ID Adds 01-features/05-authenticate-and-authorize/05-certificate-based-auth/ with: - Okta: DCR service app, JWK registration, M2M + OBO flows, browser 3LO helper - Entra ID: KMS-signed X.509 cert, service app + web app registration, M2M + OBO flows - Shared: KMS RSA_2048 signing key, offline unit tests (48 total, all passing) Feature reference: https://docs.aws.amazon.com/bedrock-agentcore/latest/devguide/private-key-jwt.html Updates parent README.md with entries in the Top-level layout table, Auth Pattern Quick Reference, Finding Things, and Running the Python Scripts sections. * Address PR review: eliminate SHA-1 usage, fix all ruff findings, correct stale folder paths CodeQL HIGH findings (SHA-1 on cert DER, 4 alerts): - Removed hashlib.sha1() usage from production code entirely - thumbprints() -> x5t_s256_thumbprint(): returns SHA-256 only - x5t (SHA-1 base64url) was never used by AgentCore's Entra provider, which always uses x5t#S256; only the display line printed it - Portal SHA-1 fingerprint (shown by Entra admin UI) is now looked up by users via `openssl x509 -in entra_cert.pem -noout -fingerprint -sha1` - Test in entra/tests/test_provider_config.py updated to test only x5t#S256 Ruff check + format: - 425 auto-fixable (F541 f-string-missing-placeholders, I001 unsorted imports, RUF100 unused noqa) - 19 manual fixes: - BLE001 (15): narrowed blind Exception catches to specific types - JWT payload decoders: (ValueError, IndexError, TypeError) - Graph HTTP calls: (requests.RequestException, ValueError, KeyError) - base64 decode: (ValueError, TypeError) - JSON parse: ValueError - cert compare: (ValueError, KeyError, TypeError) - TRY004 (3): raise TypeError (not ValueError) on non-RSA key check - S110 (1): resolved along with the co-located BLE001 narrowing - 27 files reformatted to match line-length=120 (repo pyproject.toml) Docs: - Stale folder path fix in okta/README.md and entra/README.md: the copy-paste `cd 05-outbound-auth-private-key-jwt/{okta,entra}` example was renamed to `cd 05-certificate-based-auth/{okta,entra}` Test verification: 48 unit tests pass (15 Okta + 33 Entra) offline. Remaining CodeQL "clear-text logging of sensitive information" HIGH findings on client_id printing and .env writes are false positives: - client_id is a public OAuth 2.0 identifier per RFC 6749 section 2.2 - .env writes contain only public identifiers (KMS ARN, kid, thumbprint, client_id), no secrets These should be dismissed in the Code Scanning UI with the above rationale. |
||
|
Neha Thakur
|
ff11ccbb89 |
Add Gateway, Policy Engine, Guardrails, Evaluators, and Observability to Video Games Sales Assistant (#1696)
* feat(video-games-sales-assistant): Add all AgentCore features - Gateway, Policy Engine, Guardrails, Evaluators, Observability - Add AgentCore Gateway with MCP Lambda target for PostgreSQL tools - Add Cedar Policy Engine (allow default, block PII/cost columns) - Add Bedrock Guardrail for content filtering (PII + cost topics) - Add LLM-as-a-Judge evaluators (SqlAccuracy, ResponseQuality) - Add Gateway observability (CloudWatch Logs delivery) - Add deploy.py for script-based deployment alternative - Add evaluations/evaluate.py harness with 8 test scenarios - Add setup-frontend.sh for automated frontend configuration - Expand README to 484 lines covering all AgentCore features * fix(amplify): Add monorepo amplify.yml at repo root for Amplify Hosting * fix(amplify): Add esbuild as direct devDependency for Amplify Hosting builds pnpm exec cannot find esbuild binary when it is only a transitive dependency. Adding it directly ensures the binary is linked in .bin/. Also removed --frozen-lockfile to allow platform-specific resolution. * fix(amplify): Fix Amplify Hosting build - update lockfile, skip backend deploy * fix(amplify): Generate amplify_outputs.json at build time from env vars * fix(amplify): Generate amplify_outputs.json at build time from env vars * docs: Update READMEs for Amplify Hosting, remove amplify_outputs.json from tracking * docs: Rewrote parent README as self-sufficient deploy guide, added architecture diagram * Add nehatb to CONTRIBUTORS.md * feat: Replace standalone Bedrock Guardrail with AgentCore guardrails-in-policy - Use Cedar guardrails-in-policy at Gateway Policy Engine layer - Move SQL content validation to Lambda handler - Remove standalone AWS::Bedrock::Guardrail and GUARDRAIL_ID env vars - Remove deploy-policies.sh (policies now in single CDK deploy) - Rename video-games-sales-assistant to data-analyst-conversational-assistant - Update architecture diagram * fix: Address PR review feedback - batch evals, guardrail docs, agentcore CLI reference - Replace manual scenario iteration with BatchEvaluationRunner using native Dataset management API for server-side batch evaluation - Add guardrail testing section with curl examples for prompt injection, harmful content, and PII suppression - Remove agentcore-cli tag from agentcore.json * fix: ruff formatting and missing time import in data-analyst-conversational-assistant --------- Signed-off-by: Bharathi Srinivasan <bhrsrini@amazon.com> Co-authored-by: nehatb <nehatb@amazon.com> Co-authored-by: Bharathi Srinivasan <bhrsrini@amazon.com> |
||
|
Julia Furst Morgado
|
580e6b5e00 |
feat(observability): add Dash0 as partner observability integration (#1539)
feat(observability): add Dash0 as partner observability integration Signed-off-by: Julia Furst Morgado <52685951+juliafmorgado@users.noreply.github.com> Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com> |
||
|
Guruprasad
|
423e7d7f5d |
Claude SDK and Google ADK supported-frameworks (#1772)
* feat(evaluations): add supported-frameworks samples (Google ADK, Claude Agent SDK) Add evaluation code samples for two newly-supported agent frameworks: - google-adk/ — HR Assistant built with Google ADK + LiteLLM (routes to Bedrock, no external API key). Instrumented with openinference-instrumentation-google-adk >= 0.1.13. - claude-agent-sdk/ — HR Assistant built with Claude Agent SDK. Instrumented with openinference-instrumentation-claude-agent-sdk >= 0.1.3. Both samples: - Implement the same HR Assistant (same 5 tools, shared mock data) - Deploy to AgentCore Runtime as containers - Run on-demand evaluation (Builtin.GoalSuccessRate, Correctness, Helpfulness + custom LLM-as-a-judge) - Create online evaluation config for continuous monitoring - Include idempotent cleanup scripts Shared mock_data.py extracted for reuse across framework samples. * test: add unit tests, fix Dockerfiles, add .env.example - Add 29 unit tests (mock data, tool logic, deploy/cleanup, Dockerfile validation) - Fix Dockerfiles: remove invalid COPY ../ (use in-context shared/ copy) - Add .env.example for both samples (template, no secrets) - Validates: no real PII, tool correctness, eval scenario consistency * fix(claude-agent-sdk): correct SDK API usage after local verification Claude Agent SDK (v0.2.115) uses ClaudeSDKClient + MCP tools, not a simple Agent class. Rewritten to use: - create_sdk_mcp_server() for tool registration - @tool decorator with MCP-style input_schema - ClaudeSDKClient with send_message() for invocation - ResultMessage for response extraction Verified locally: all imports resolve correctly for both google-adk and claude-agent-sdk. * docs: add Guruprasad Seeryada to CONTRIBUTORS.md * fix: remove f-string without placeholders (ruff F541) * style: apply ruff format to all Python files * fix: address reviewer feedback (Bharathi) - Remove tests/ folder (reviewer request) - Add AgentCore eval support explanation + docs links to both READMEs - Add example trace JSON for both frameworks (AGENT, TOOL, LLM spans) - Replace ASCII architecture diagrams with draw.io source files (export as PNG before merge — reviewer requested hand-drawn diagrams) - READMEs now reference architecture.png (to be exported from .drawio) * docs: add architecture diagrams (AWS icons, draw.io export) --------- Signed-off-by: Bharathi Srinivasan <bhrsrini@amazon.com> Signed-off-by: Guruprasad <116177025+svguruprasad@users.noreply.github.com> Co-authored-by: Bharathi Srinivasan <bhrsrini@amazon.com> |
||
|
Massimiliano Angelino
|
879b67cbbf |
feat(gateway-ide): gateway session binding and DCR example (#1758)
* chore: example that shows how to add DRC and automatic session binding to AgentCore Gateway also shows how to add targets that themselves only support DCR * fix: ruff and bandit fixes * fix(secure-ide-gateway-tool): resolve ASH findings - bandit B310: move nosec onto urlopen line after https scheme validation - detect-secrets: annotate demo credentials with pragma allowlist (utils.py, notebook) - grype: bump aws-cdk-lib to ^2.261.0 (CVE fix) and migrate off @aws-cdk/aws-bedrock-agentcore-alpha to stable aws-cdk-lib/aws-bedrockagentcore - remove unused jest tooling (no tests) to drop transitive js-yaml vuln - standardize on single pnpm-lock.yaml * style: apply ruff format to secure-ide-gateway-tool python files * chore: restore 03-ide-gateway-tool demo video and stop tracking SECURITY-REVIEW.md * doc: fixed README and DEPLOYMENT * fix: masked ids * fix: removed secrets, added diagrams |
||
|
Anil Nadiminti
|
496c79e72b | feat(agents-that-transact): use LangGraph payments middleware (AgentCorePaymentsMiddleware) (#1783) | ||
|
Shreya Pawaskar
|
5c26a2e629 |
Add 04-fmkb-managed-kb sample (Managed KB via AgentCore Gateway) (#1770)
* Add 04-fmkb-managed-kb sample to connect-your-agent-to-anything Exposes a Bedrock Managed Knowledge Base (FMKB) as an MCP tool via AgentCore Gateway, queried by a Strands agent on AgentCore Runtime. Includes a raw MCP/SigV4 path (01-raw-mcp) and the deployed-agent path (02-strands-agent), plus KB + gateway helpers. * Replace ASCII diagram with AWS-icon architecture image Adds images/architecture.png (official AWS Bedrock/IAM icons) and embeds it in the README in place of the ASCII box drawing. * style: apply ruff format to satisfy Python Code Quality CI Reformats the sample's Python files to the repo's ruff config (line-length=120). Cosmetic only — no logic changes. * Update architecture diagram Refreshes images/architecture.png with cleaner node labels and edge-centered labels (white background). --------- Co-authored-by: shraiyya <shraiyya@users.noreply.github.com> |
||
|
Renya Kujirada
|
2ac6a0986b |
feat(evaluations): add supported-frameworks samples for OpenAI Agents SDK and LlamaIndex (#1774)
* feat(evaluations): add supported-frameworks samples for OpenAI Agents SDK and LlamaIndex Add one evaluation code sample per newly-supported framework under 02-evaluate/supported-frameworks/. Each sample re-implements the shared HR Assistant, deploys it to AgentCore Runtime with OpenTelemetry instrumentation (auto-discovered by ADOT), and evaluates it with built-in and custom LLM-as-a-judge evaluators (on-demand and online). - openai-agents: gpt-oss via the Bedrock mantle Responses API endpoint (OpenAIResponsesModel + short-term bearer token from the runtime role) - llamaindex: FunctionAgent workflow with BedrockConverse (Nova Lite), streaming disabled for clean inference spans - both: conversation history persisted in AgentCore Memory per session, telemetry force-flushed before the microVM freezes Verified end-to-end on us-west-2: deploy, multi-turn invocation with cross-turn references, on-demand evaluation (11/11 valid scores), and online evaluation results in CloudWatch. * docs(evaluations): expand supported-frameworks READMEs for education - Add 'What you'll learn' concept tables and architecture diagrams to both framework sample READMEs, matching the sibling samples' style - Add 'Expected output' sections with real run transcripts so learners can verify their results - Add 'Evaluate from the CLI' sections using the AgentCore CLI's standalone mode (agentcore run eval --runtime-arn/--evaluator-arn), verified against the deployed runtimes - Add a 'Making any framework agent evaluable' recipe and next-steps links to the folder README, generalizing the integration steps * feat(evaluations): switch OpenAI Agents sample to GPT-5.5 with Bedrock API keys Replace gpt-oss-120b with openai.gpt-5.5, served on the mantle endpoint's openai/v1 path from us-east-1/us-east-2 (the runtime calls it cross-region via BEDROCK_OPENAI_MODEL_REGION). Authentication now follows the Bedrock API key model: a short-term key is minted from the runtime's IAM role on every invocation (secure default), and a long-term key can be supplied via the BEDROCK_API_KEY environment variable. README documents both key types and how to generate a long-term key. Verified end-to-end on the deployed runtime: 3-turn session with cross-turn references, AgentCore Memory persistence, and on-demand evaluation returning 11/11 valid scores (GoalSuccessRate 1.0, Correctness 1.0, HRSessionCompleteness 1.0). * refactor(evaluations): simplify OpenAI auth to provide_token directly provide_token() already returns a short-term Bedrock API key, so the BEDROCK_API_KEY environment-variable branch was unnecessary. Call it directly in _build_agent and drop the _get_api_key helper. README now documents the long-term key as a pass-it-as-api_key alternative rather than a separate env-var code path. Behavior is identical to the deployed runtime's default path (verified by smoke-testing the live GPT-5.5 runtime). * docs(evaluations): rename abrupt 'Why the same HR Assistant?' section Replace the question-style heading with 'The shared HR Assistant scenario' and move it right after the Samples table, where the HR Assistant is first mentioned, so the README flows from what the samples are to why they share one scenario. * fix(evaluations): suppress spurious pip warnings during package build The deployment zip is built with pip install --target into an isolated directory, but pip still checks the resolved set against packages in the ambient environment and prints misleading 'ERROR: pip's dependency resolver...' conflicts that do not affect the deployed artifact. Add --no-warn-conflicts and --disable-pip-version-check so deploy.py output only shows real errors. * fix(evaluations): address supported framework review feedback * fix(evaluations): correct image reference and ruff formatting |
||
|
Anil Nadiminti
|
eda4f2bb09 |
docs(08-agents-that-transact): convert payments getting-started to AgentCore CLI + SDK (#1767)
--------- Co-authored-by: Anil Nadiminti <anilnadi@amazon.com> Co-authored-by: deepaxs <deepaxs@amazon.com> |
||
|
Fabio Balancin
|
790838a830 |
Add egress-controlled code execution sample (#1769)
* Add egress-controlled code execution sample A multi-container security sample under 03-advanced: a trusted supervisor container (started by AgentCore Runtime) launches a broker and an untrusted agent container inside the same microVM via containerd. All agent egress is mediated by the broker, which enforces a runtime-configurable allowlist. The Phase 1 operation is ping_domain, exercising the full supervisor -> agent -> broker -> egress path. The allow-vs-deny contrast (amazon.com permitted, aws.amazon.com denied) demonstrates the security boundary end to end. Includes deploy.py / invoke.py / cleanup.py (matching the folder convention), three arm64 Dockerfiles, unit tests for the IPC framing and broker allowlist, and an architecture diagram. Also adds the row to the 03-advanced index. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * Add contributors to CONTRIBUTORS.md Add Fabio Balancin (balancin) and Varun Gunda (vvargu) for the egress-controlled code execution sample. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * Remove redundant Roadmap and Cost Considerations sections from egress-coding-execution README Consolidate into the existing Current status section and drop the broken Roadmap anchor link. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Fabio Balancin <balafabi@amazon.com> Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com> |
||
|
Bharathi Srinivasan
|
dbe1c81bd6 |
Remove comments and default configuration from dependabot.yml (#1776)
Signed-off-by: Bharathi Srinivasan <bhrsrini@amazon.com> |
||
|
Bharathi Srinivasan
|
473df951da |
Remove no-op Dependabot metadata workflow (#1775)
.github/workflows/dependabot.yml only fetched PR metadata from dependabot[bot] PRs but had no downstream steps wired up, making it a dead workflow. Removing to reduce CI noise. |
||
|
Renya Kujirada
|
366816c0e3 |
Add Policy in Amazon Bedrock AgentCore samples: tool access control and guardrails in policies (with architecture diagrams) (#1768)
* Add AgentCore Policy samples: tool access control and guardrails as policies * fix: apply ruff formatting to policy sample Python files * docs: rename AgentCore Policy to "Policy in Amazon Bedrock AgentCore"; rename 03-guardrails-as-policies to 02-guardrails-as-policies * update readme * fix: remove invalid --target flag from agentcore add policy CLI commands; add --form-data-path; update install to @latest * docs(policy): add image-based architecture diagrams to policy READMEs Replace ASCII architecture diagrams with draw.io generated PNG diagrams using official AWS icons (Bedrock AgentCore Gateway, AgentCore Policy, Lambda, Cognito, Bedrock Guardrails) in: - 02-policy/README.md - 02-policy/01-tool-access-with-policy/README.md - 02-policy/02-guardrails-in-policy/README.md The PNGs embed the draw.io diagram XML, so they remain fully editable by opening them in draw.io. * docs(policy): remove 'Cedar' wording from architecture diagrams Per reviewer feedback: the authorization language is no longer strictly Cedar, so refer to the policies simply as 'policies' across all three diagrams (main policy, tool-access, guardrails). * docs(policy): rename guardrails diagram title to Amazon Bedrock Guardrails Per reviewer feedback on PR #1768: refer to Guardrails in the title as Amazon Bedrock Guardrails. --------- Co-authored-by: Bharathi Srinivasan <bhrsrini@amazon.com> |
||
|
satveerkhurpa
|
8605cff717 |
Fix broken example links in obo-training OBO Reference Guide (#1763)
The Runnable Examples section had relative links to examples/ which was renamed to 3-examples/ before the initial merge. Also generalize the walkthrough paragraph so it covers both Entra and Okta claim conventions (the merged version was Entra-specific). - Relative links now correctly point at 3-examples/* - Marks UC2 Okta complete (was 'Okta planned') - Generalizes the walkthrough: actor claim (azp/appid vs cid), identity claim (oid vs sub+uid), and audience behavior (rotates on Entra vs constant on Okta) Co-authored-by: Satveer Khurpa <khurpas@amazon.com> |
||
|
Will Matos
|
69fb6dd640 |
Enhance event-driven-claims-agent: cost routing, deterministic execution, docs (#1761)
* feat(usecases): it-incident-response
* add jira integration
* fix minor memory usage points
* docs: add IT incident response agent project docs and assets
* config: add AgentCore project configuration and schema context
* infra: add CDK project for AgentCore L3 construct deployment
* feat: add Strands agent application (runtime, memory, MCP client, model)
* feat: add Lambda functions for tools, infra providers, and ticket trigger
* feat: add tool schemas, seed data, and knowledge base runbooks
* chore: add deployment, evaluation, and ticket utility scripts
* docs: comply with use-case README template (add details table and disclaimer)
* chore: allow esbuild install script in CDK project
* fix: create S3 Vectors bucket+index for KB (CFN does not auto-create)
The Bedrock KnowledgeBase CloudFormation resource with type S3_VECTORS
requires a pre-existing vector bucket and index. Passing an empty
s3VectorsConfiguration fails schema validation. The console's 'quick
create' auto-provisions these, but CloudFormation does not.
Changes:
- Import aws-cdk-lib/aws-s3vectors
- Create CfnVectorBucket (named per account+region)
- Create CfnIndex (float32, 1024 dims, cosine, no metadata keys)
- Wire IndexArn, IndexName, VectorBucketArn into KB storageConfiguration
- Grant KB role s3vectors:* actions on the vector index
- Add removalPolicy to bucket+index
- Add troubleshooting row to README
- Deleted the ROLLBACK_COMPLETE stack to unblock next deploy
* fix: resolve deploy issues (template envsubst, target name, IndexName removal)
- aws-targets.json.template: fix envsubst-incompatible default syntax
(${AWS_REGION:-us-west-2} → ${AWS_REGION}), change name 'dev' → 'default'
- scripts/deploy.sh: export AWS_REGION with default before envsubst runs
- infra-construct.ts: remove IndexName from s3VectorsConfiguration (caused
CFN 'oneOf 2 subschemas matched' validation error — only IndexArn +
VectorBucketArn are needed for BYO S3 Vectors)
- deployed-state.json: updated by successful deploy
* docs: update Quickstart to lead with deploy.sh as primary path
* chore: stop tracking deployed-state.json (deployment-specific, not shared)
* refactor(auth): rename OAUTH_PROVIDER_NAME to GATEWAY_OAUTH_PROVIDER_NAME
Scope auth env vars per boundary for clarity:
- OAUTH_PROVIDER_NAME -> GATEWAY_OAUTH_PROVIDER_NAME
- GATEWAY_AUDIENCE -> GATEWAY_OAUTH_AUDIENCE
Code maintains backward-compat fallback to legacy names.
CDK injects new names into Runtime env vars at deploy.
.env.example and agentcore/.env.local.example updated
with boundary labels (Boundary 2, Boundary 3).
* docs(auth): add authentication boundary guide
New docs/authentication-guide.md explains the 3 auth boundaries:
1. Runtime Inbound (SigV4 vs CUSTOM_JWT)
2. Gateway Outbound (AWS_IAM vs CUSTOM_JWT M2M)
3. Jira Outbound (USER_FEDERATION 3LO)
Covers: conceptual overview, env var reference, local dev
implications, troubleshooting, and why all 3 patterns exist.
* docs: fix local dev ports, CLI commands, and env var references
README.md:
- Add Port Mapping section (8081=Web UI, 8082=Runtime container)
- Add troubleshooting entry for workload access token error
- Fix agentcore invoke --dev (invalid) -> agentcore dev prompt
- Update config table to use GATEWAY_OAUTH_* var names
docs/custom-jwt-auth-upgrade.md:
- Update all OAUTH_PROVIDER_NAME -> GATEWAY_OAUTH_PROVIDER_NAME
- Update all GATEWAY_AUDIENCE -> GATEWAY_OAUTH_AUDIENCE
docs/ARCHITECTURE.md:
- Wrap ASCII diagrams in <details> tags for readability
* Fix S3 Vectors Knowledge Base schema: remove indexName from s3VectorsConfiguration
CloudFormation's AWS::Bedrock::KnowledgeBase type with S3_VECTORS storage
has a schema constraint (oneOf) that rejects when both indexArn and indexName
are provided. Pass only vectorBucketArn + indexArn to satisfy the schema.
The indexName is implicit in the indexArn and Bedrock manages metadata
internally during ingestion, so it's not needed in the configuration.
Fixes: Properties validation failed - 'only 1 subschema matches out of 2'
Verified: Stack now deploys successfully with CREATE_COMPLETE status
* feat: migrate online eval to declarative agentcore.json
Remove the custom resource workaround for Online Evaluation. The
AgentCoreOnlineEvaluationConfig L3 construct (@aws/agentcore-cdk
v0.1.0-alpha.34+) now handles dependency ordering automatically.
Changes:
- Delete lambdas/infra/online_eval_provider.py (custom resource Lambda)
- Remove SKIP_ONLINE_EVAL logic from cdk-stack.ts and bin/cdk.ts
- Online eval is now purely declarative via agentcore.json onlineEvalConfigs[]
- To disable: set onlineEvalConfigs to [] in agentcore.json
Standards-Consulted: std.cdk.prefer-declarative
Standards-Gaps: none
Standards-Proposed: none
* fix: align model IDs with available Bedrock models
Update all model ID references to use the exact identifiers from
`aws bedrock list-foundation-models`:
- AGENT_MODEL_ID: us.anthropic.claude-sonnet-4-6 (not -20250929-v1:0)
- FAST_MODEL_ID: us.anthropic.claude-3-5-haiku-20241022-v1:0
- JUDGE_MODEL_ID: us.anthropic.claude-sonnet-4-6
The previous IDs (with -20250929-v1:0 suffix) were invalid and caused
runtime ValidationException on agent invocation.
Standards-Consulted: std.config.model-id-consistency, std.bedrock.verify-model-id-format
Standards-Gaps: none
Standards-Proposed: std.config.model-id-consistency, std.bedrock.verify-model-id-format, std.bedrock.validate-model-before-deploy
* fix: add --target dev to all agentcore CLI commands
This project uses a named target 'dev' in aws-targets.json (not the
default target). All CLI invocations must explicitly pass --target dev.
Standards-Consulted: std.deploy.target-dev
Standards-Gaps: none
Standards-Proposed: std.deploy.target-dev
* docs: overhaul documentation to describe current state only
- Remove SKIP_ONLINE_EVAL references throughout
- Fix model IDs in all documentation
- Remove .kiro references from public docs
- Merge duplicate Online Eval sections in README
- Fix GUARDRAIL_ID default (auto-creates, not skips)
- Fix target name in troubleshooting (dev, not default)
- Add --target dev to all documented deploy commands
- Add Declarative vs Imperative section to ARCHITECTURE.md
- Remove stale development process docs (8 root-level .md files)
- Pad all README tables for aligned vertical bars
Standards-Consulted: std.docs.current-state-only, std.deploy.target-dev, std.config.model-id-consistency
Standards-Gaps: none
Standards-Proposed: std.docs.current-state-only
* chore: update agentcore config and CDK dependencies
- agentcore.json: add onlineEvalConfigs, policyEngines, gateway config
- aws-targets.json.template: minor format fix
- CDK packages: update @aws/agentcore-cdk dependency range
Standards-Consulted: std.cdk.prefer-declarative, std.deploy.target-dev
Standards-Gaps: none
Standards-Proposed: none
* feat: improve agent resilience and MCP client handling
- main.py: add graceful degradation when MCP tools unavailable,
safe fallback to LLM-only mode on tool initialization failure
- mcp_client/client.py: add get_all_mcp_clients_safe() with error
collection instead of hard failure
- trigger: minor fix
- show_ticket.sh: minor fix
Standards-Consulted: std.agentcore.mcp-sigv4-auth
Standards-Gaps: none
Standards-Proposed: none
* feat: add end-to-end test script
scripts/test-e2e.sh publishes a ticket to SNS, polls DynamoDB for
resolution (10s intervals, 120s timeout), and asserts status=Resolved
with a non-empty resolution_comment. Exits 0 on pass, 1 on fail.
Usage:
./scripts/test-e2e.sh # sample ticket
./scripts/test-e2e.sh /path/to/ticket.json # custom ticket
Standards-Consulted: std.deploy.target-dev
Standards-Gaps: none
Standards-Proposed: none
* feat: upgrade FAST_MODEL_ID to Claude Haiku 4.5
Replace claude-3-5-haiku-20241022-v1:0 (legacy, access-gated) with
claude-haiku-4-5-20251001-v1:0 (current, available in account).
Verified: model invocable, E2E test passes all 3 tiers (LOW/HIGH/CRITICAL).
Standards-Consulted: std.config.model-id-consistency, std.bedrock.verify-model-id-format, std.bedrock.validate-model-before-deploy
Standards-Gaps: none
Standards-Proposed: none
* fix: add .gitignore with lib/ un-ignore for CDK source files
The root .gitignore excludes lib/ globally (for compiled JS output in
other samples). This project's CDK TypeScript source lives in
agentcore/cdk/lib/ and must be tracked. Add project-level .gitignore
with negation rules so 'git add' works without -f flag.
Standards-Consulted: std.git.no-deployed-state
Standards-Gaps: none
Standards-Proposed: none
* docs: remove sample-level LICENSE, inherit from monorepo root
Remove the local MIT-0 LICENSE file from it-incident-response-agent.
The sample should inherit the root repo's Apache 2.0 license, consistent
with all other samples in 02-use-cases/.
* feat: migrate observability to declarative agentcore.json, add Cedar policy steering
- Move OTEL/X-Ray env vars to agentcore.json runtimes[].envVars[]
- Add instrumentation.enableOtel: true
- Add onlineEvalConfigs[] (4 built-in evaluators, 100% sampling)
- Add policyEngines[] with Cedar policies (resource is AgentCore::Gateway)
- Add policyEngineConfiguration to gateway (LOG_ONLY mode)
- Remove ~150 lines of imperative CDK (custom resource, env var overrides)
- L3 construct now handles gateway lambda:InvokeFunction automatically
- Rename target from 'default' to 'dev'
- Fix trigger Lambda runtimeSessionId length (min 33 chars)
- Add Cedar policy syntax steering file
- Update README with CLI commands for online-eval + policy-engine
* chore: remove .kiro/ from git tracking, add to .gitignore
* fix: update evaluate.py
* refactor: simplify evaluate.py to retrieve online eval results
Replace complex on-demand evaluation with a script that queries the
online evaluation results log group. The continuous online evaluation
(agentcore.json onlineEvalConfigs[]) scores all invocations automatically.
Standards-Consulted: std.docs.current-state-only
Standards-Gaps: none
Standards-Proposed: none
* chore: clean up developer-only artifacts and reduce consumer confusion
- Remove AGENTS.md from tracking (Kiro AI context, not for consumers)
- Reorganize .gitignore with categories, add developer-only file exclusions
- Rename docs/online-eval-workaround.md → online-evaluation.md
- Trim ARCHITECTURE.md auth deep-dive (link to authentication-guide.md)
- Simplify README: collapse manual path, clarify CLI-first section,
remove duplicate env-var table, consolidate Configure section
* feat: add OTEL span attributes, tool-call hooks, e2e test, and eval reporter
- Add ticket.id/priority/requester_id/mode as OTEL span attributes for
end-to-end trace correlation via CloudWatch Transaction Search
- Add Strands BeforeToolCallEvent/AfterToolCallEvent hooks for per-tool
call timing in runtime logs
- Add scripts/e2e_test.py with live status polling, log tailing, and
post-resolution tool call timeline
- Rewrite scripts/evaluate.py to parse online eval results with summary
(avg scores by evaluator) and detailed per-trace breakdown
* remove: deprecate 02-use-cases/it-incident-response-agent (v1)
The v1 sample at 02-use-cases/it-incident-response-agent/ is superseded by
02-use-cases/automation-agents/it-incident-response-agent/, which is an
evolved version of the same use case with significant improvements:
- CLI-first workflow (agentcore.json + agentcore deploy) vs raw CDK
- Zero external prerequisites to deploy (Auth0/Jira optional, not required)
- All 6 AgentCore services demonstrated (adds Policy Engine, Guardrails)
- Production patterns: DLQ, idempotency, cost routing, graceful degradation
- Local dev support (agentcore dev with hot-reload)
- L3 CDK constructs instead of L1 CfnResource boilerplate
- Comprehensive documentation and design-decisions ADRs
The v1 code used raw L1 CfnResource constructs with mandatory Auth0 +
Jira dependencies, making it inaccessible for quick-start consumers.
All v1 functionality (Jira integration, Auth0 CUSTOM_JWT, Atlassian 3LO,
online evaluation) is preserved in the automation-agents version as
optional toggles.
* style: fix lint and format issues for CI compliance
Python (ruff):
- Remove unused imports: get_all_mcp_clients, get_streamable_http_mcp_client
(main.py), os (jira_oauth_provider.py), time (seeder.py)
- Auto-format 8 files to pass ruff format check
TypeScript (Prettier):
- Auto-format cdk-stack.ts, infra-construct.ts, bin/cdk.ts
All CI checks now pass:
- ruff check: 0 errors
- ruff format --check: 21 files formatted
- prettier --check: all files pass
- tsc --noEmit: compiles clean
- agentcore validate: Valid
* chore: gitignore eval doc build artifacts
* Potential fix for pull request finding 'CodeQL / Clear-text logging of sensitive information'
Co-authored-by: Copilot Autofix powered by AI <62310815+github-advanced-security[bot]@users.noreply.github.com>
Signed-off-by: Will Matos <wilmatos@amazon.com>
* Potential fix for pull request finding 'CodeQL / Clear-text logging of sensitive information'
Co-authored-by: Copilot Autofix powered by AI <62310815+github-advanced-security[bot]@users.noreply.github.com>
Signed-off-by: Will Matos <wilmatos@amazon.com>
* fix: resolve ruff lint errors in e2e_test.py
- F541: Remove extraneous f-prefix on string without placeholders (line 157)
- F841: Remove unused variable base_ts (line 270)
Standards-Consulted: std.python.ruff-lint-clean
Standards-Gaps: none
Standards-Proposed: none
* fix: additional e2e_test.py improvements
* fix: e2e_test.py refinements
* fix: code review improvements - error handling, threading, cleanup
- Wrap _resolve_ticket DDB call in try/except (non-fatal failure)
- Add DEBUG log for silent OTEL ImportError
- Remove unused imports in mcp_client/jira.py
- Add docstrings to memory/session.py
- Move import threading to top-level in e2e_test.py
- Replace mutable list stop flag with threading.Event
Verified: all Python files parse cleanly, tsc --noEmit passes,
agentcore validate passes, CDK tests pass, 50-ticket E2E test 100% pass.
* feat: enable Transaction Search observability and tighten IAM scoping
- Add OTEL/GenAI observability env vars (AGENT_OBSERVABILITY_ENABLED,
message-content capture, application signals) to agentcore.json
- Add Transaction Search custom resource Lambda to route X-Ray segments
to CloudWatch Logs (aws/spans) for online eval ingestion
- Scope trigger Lambda to bedrock-agentcore:InvokeAgentRuntime on the
specific runtime ARN (least privilege)
- Add resource-based lambda:InvokeFunction permissions for Gateway targets
- Make _fail_ticket DDB write non-fatal with exception logging
- Misc lint/cleanup in lambdas and e2e_test.py
* docs: reflect auto-enabled Transaction Search via custom resource
Transaction Search is now provisioned automatically by the CDK stack
(transaction_search.py custom resource) when onlineEvalConfigs is set,
so manual 'aws application-signals start-monitoring' is no longer required.
Update README, online-evaluation.md, and ARCHITECTURE.md accordingly.
* refactor: remove redundant L3-provided config, restore Memory resource
- Remove redundant XRayTracing IAM statement from RuntimeAdditionalPolicy (L3 RuntimeExecutionRole already grants xray put-trace perms); keep CloudWatch Logs Insights
- Remove instrumentation.enableOtel from agentcore.json (OTEL wrapping provided by Dockerfile CMD opentelemetry-instrument for Container build)
- Move AGENT_MODEL_ID/FAST_MODEL_ID into agentcore.json runtimes[].envVars[] (declarative, was imperative addPropertyOverride)
- Restore Memory resource (ITIncidentAgentMemory, SUMMARIZATION, namespace incidents/{actorId}) so memory code is backed by a provisioned resource instead of a no-op
- Enable ACTIVE X-Ray tracing on trigger Lambda for full service-map coverage
- Update README/ARCHITECTURE/online-evaluation docs; add 'agentcore add memory' CLI instructions
* docs: fix stale references in auth and schema docs
- authentication-guide: JIRA_MCP_URL is a hardcoded constant set by CDK, not auto-derived from JIRA_SITE_URL
- custom-jwt-auth-upgrade: correct stale function name _get_oauth_token() -> _create_custom_jwt_client()
- .llm-context/README: remove reference to non-existent mcp.ts schema file
* fix: deploy blockers for Memory namespace and gateway target IAM ordering
- Memory SUMMARIZATION namespace requires {sessionId}: incidents/{actorId} -> incidents/{actorId}/{sessionId} (CreateMemory validation failed without it). Retrieval still works via the incidents/{actorId} prefix (prefix matching).
- GatewayTargets now explicitly depend on the gateway role DefaultPolicy. The AgentCoreMcp L3 creates the lambda:InvokeFunction policy but does not order targets after it, causing deterministic 'Gateway execution role lacks permission' CREATE_FAILED. Replaces the ineffective resource-based fn.addPermission approach.
- README: Getting Started section (CI doc gate), Memory namespace + sessionId note
- ARCHITECTURE: corrected Memory namespace note
* style: lint fixes (ruff/pylint) with no behavior change
- main.py: remove unused imports GATEWAY_URL, MEMORY_ID (ruff F401); they are imported where actually used in mcp_client/memory modules
- mcp_client/client.py: remove unnecessary else-after-return (pylint R1705)
- mcp_client/client.py, jira.py: scoped 'pylint: disable=missing-kwoa' with justification on @requires_access_token-decorated calls (decorator injects access_token; pylint cannot see the transform)
* style: black formatting in lambda handlers (no behavior change)
- transaction_search.py: collapse wrapped log line
- ticket_event_handler.py: expand ALLOWED_FIELDS set to multi-line, collapse RuntimeError
* fix: auto-trigger KB ingestion on deploy by passing DataSourceId to seeder
The seeder gates ingestion on 'kb_id and data_source_id', but the CDK only passed KnowledgeBaseId — DataSourceId was never wired through, so start_ingestion_job never ran and the auto-created KB stayed empty until manual ingestion.
Capture the data source id on InfraConstruct (knowledgeBaseDataSourceId) and pass DataSourceId to the TriggerSeeder custom resource. README updated: ingestion is now automatic; manual command kept as optional re-index.
* fix: validate ticket_id/issue_key in agent entrypoint to avoid uncaught KeyError
payload['ticket_id'] was read outside the main try block, so a malformed/direct invoke missing ticket_id (and issue_key) raised an uncaught KeyError that crashed the invoke generator with no structured response. Use payload.get() with an explicit guard that yields a structured Failed result and returns; tighten is_jira_mode to bool(JIRA_MCP_URL).
* fix: agent entrypoint robustness
Guard model output extraction; empty output raises ValueError so the ticket is marked Failed (requires human processing) rather than resolved with a fallback. Sanitize both title and description through the guardrail. Remove dual logger alias. Avoid unbound agent reference when both initializations fail in prompt mode.
Findings #3, #6, #7, #8
* fix: memory cross-session retrieval
Use namespace_path (prefix match) instead of namespace (exact match) so prior-session memories are retrieved during enrichment.
Finding #4
* fix: atomic change-request write + reason field
Use transact_write_items for the dual DynamoDB write so the change request and audit records commit atomically. Add the reason parameter to the tool schema and wire it through, making the RequireReasonForChangeRequest Cedar policy functional.
Findings #5, #12
* fix: DynamoDB Decimal serialization in tools
Convert DynamoDB Decimal values to native int/float before json.dumps so tool responses serialize correctly.
Finding #15
* fix: trigger region + KB ingestion re-run + QueryKb env
Pass region_name to boto3 clients in the ticket event trigger. Add commonEnv to QueryKbFn. Bump the seeder Version to 3 so the KB data source ingestion re-runs.
Findings #9, #10, #14
* fix: e2e span duration unit
Correct the span duration unit conversion in the e2e test.
Finding #11
* style: pylint cleanups in e2e_test and evaluate scripts
- Drop unused loop variable, specify UTF-8 encoding on open()
- Simplify elif-after-return in _score_bar
- Add main() docstrings and wrap long lines
* style: wrap long lines in agent and lambda modules (C0301)
* style: apply ruff format (line-length 120)
* chore: rename folder
* feat(docs): update docs
* chore(usecases) - restructure under workflow
* feat(it-incident-agent): stream real-time pipeline stages + evaluations UI + demo
- Agent (main.py): emit real SSE stage events at each pipeline phase
(guardrail, memory, tools, diagnose, per-tool-call, persist, emit)
using agent.stream_async() instead of blocking agent() call
- Docs: add Real-Time Progress Streaming section to ARCHITECTURE.md
- README: add demo GIF showcasing the full workflow
- lambdas/tools: minor cleanup in create_change_request.py
- .pylintrc: project lint config
* style: add blank line before _stage_event function
* chore: update .gitignore for AgentCore CLI and CDK artifacts
* feat: add AgentCore CDK infrastructure (DynamoDB, Lambda, Cognito, EventBridge)
* feat: implement dual-agent claims processor with memory and MCP Gateway
* feat: update Lambda tool handlers with input validation and routing
* feat: add one-command deploy/destroy scripts and .env.example
* feat: add E2E test suite, Cedar tests, lint script, and unit tests
* docs: add architecture, deployment guide, ADRs, and update README
* refactor: remove legacy infra/ directory (replaced by agentcore/cdk/)
* chore: add gitignore negations for claims-agent CDK lib and Dockerfile
* fix: switch test_invoke.py to SigV4 auth (Runtime uses AWS_IAM, not JWT)
* docs: regenerate architecture diagrams and fix Runtime auth description
* style: fix import sorting and trailing whitespace (ruff auto-fix)
* chore: add wilmatos to CONTRIBUTORS.md
* style(event-driven-claims-agent): apply ruff format
* chore(event-driven-claims-agent): improve lint.sh with verbose output, format check, and tsc
* fix(claims-agent): use SigV4 for Runtime invocation, not JWT
The Runtime uses IAM (SigV4) auth; CUSTOM_JWT is only for the Gateway.
- Trigger Lambda: replace Cognito JWT flow with SigV4 signing
- test_e2e.py: switch from Bearer token to SigV4, relax assertions
to match actual agent streaming output format
- Dockerfile: add non-root user and healthcheck (CKV_DOCKER_2/3)
- test_local.py: add URL scheme validation, suppress S310 lint
* feat(it-incident-response-agent): production-ready IT incident response agent with streaming and observability (#1724)
* feat(usecases): it-incident-response
* add jira integration
* fix minor memory usage points
* docs: add IT incident response agent project docs and assets
* config: add AgentCore project configuration and schema context
* infra: add CDK project for AgentCore L3 construct deployment
* feat: add Strands agent application (runtime, memory, MCP client, model)
* feat: add Lambda functions for tools, infra providers, and ticket trigger
* feat: add tool schemas, seed data, and knowledge base runbooks
* chore: add deployment, evaluation, and ticket utility scripts
* docs: comply with use-case README template (add details table and disclaimer)
* chore: allow esbuild install script in CDK project
* fix: create S3 Vectors bucket+index for KB (CFN does not auto-create)
The Bedrock KnowledgeBase CloudFormation resource with type S3_VECTORS
requires a pre-existing vector bucket and index. Passing an empty
s3VectorsConfiguration fails schema validation. The console's 'quick
create' auto-provisions these, but CloudFormation does not.
Changes:
- Import aws-cdk-lib/aws-s3vectors
- Create CfnVectorBucket (named per account+region)
- Create CfnIndex (float32, 1024 dims, cosine, no metadata keys)
- Wire IndexArn, IndexName, VectorBucketArn into KB storageConfiguration
- Grant KB role s3vectors:* actions on the vector index
- Add removalPolicy to bucket+index
- Add troubleshooting row to README
- Deleted the ROLLBACK_COMPLETE stack to unblock next deploy
* fix: resolve deploy issues (template envsubst, target name, IndexName removal)
- aws-targets.json.template: fix envsubst-incompatible default syntax
(${AWS_REGION:-us-west-2} → ${AWS_REGION}), change name 'dev' → 'default'
- scripts/deploy.sh: export AWS_REGION with default before envsubst runs
- infra-construct.ts: remove IndexName from s3VectorsConfiguration (caused
CFN 'oneOf 2 subschemas matched' validation error — only IndexArn +
VectorBucketArn are needed for BYO S3 Vectors)
- deployed-state.json: updated by successful deploy
* docs: update Quickstart to lead with deploy.sh as primary path
* chore: stop tracking deployed-state.json (deployment-specific, not shared)
* refactor(auth): rename OAUTH_PROVIDER_NAME to GATEWAY_OAUTH_PROVIDER_NAME
Scope auth env vars per boundary for clarity:
- OAUTH_PROVIDER_NAME -> GATEWAY_OAUTH_PROVIDER_NAME
- GATEWAY_AUDIENCE -> GATEWAY_OAUTH_AUDIENCE
Code maintains backward-compat fallback to legacy names.
CDK injects new names into Runtime env vars at deploy.
.env.example and agentcore/.env.local.example updated
with boundary labels (Boundary 2, Boundary 3).
* docs(auth): add authentication boundary guide
New docs/authentication-guide.md explains the 3 auth boundaries:
1. Runtime Inbound (SigV4 vs CUSTOM_JWT)
2. Gateway Outbound (AWS_IAM vs CUSTOM_JWT M2M)
3. Jira Outbound (USER_FEDERATION 3LO)
Covers: conceptual overview, env var reference, local dev
implications, troubleshooting, and why all 3 patterns exist.
* docs: fix local dev ports, CLI commands, and env var references
README.md:
- Add Port Mapping section (8081=Web UI, 8082=Runtime container)
- Add troubleshooting entry for workload access token error
- Fix agentcore invoke --dev (invalid) -> agentcore dev prompt
- Update config table to use GATEWAY_OAUTH_* var names
docs/custom-jwt-auth-upgrade.md:
- Update all OAUTH_PROVIDER_NAME -> GATEWAY_OAUTH_PROVIDER_NAME
- Update all GATEWAY_AUDIENCE -> GATEWAY_OAUTH_AUDIENCE
docs/ARCHITECTURE.md:
- Wrap ASCII diagrams in <details> tags for readability
* Fix S3 Vectors Knowledge Base schema: remove indexName from s3VectorsConfiguration
CloudFormation's AWS::Bedrock::KnowledgeBase type with S3_VECTORS storage
has a schema constraint (oneOf) that rejects when both indexArn and indexName
are provided. Pass only vectorBucketArn + indexArn to satisfy the schema.
The indexName is implicit in the indexArn and Bedrock manages metadata
internally during ingestion, so it's not needed in the configuration.
Fixes: Properties validation failed - 'only 1 subschema matches out of 2'
Verified: Stack now deploys successfully with CREATE_COMPLETE status
* feat: migrate online eval to declarative agentcore.json
Remove the custom resource workaround for Online Evaluation. The
AgentCoreOnlineEvaluationConfig L3 construct (@aws/agentcore-cdk
v0.1.0-alpha.34+) now handles dependency ordering automatically.
Changes:
- Delete lambdas/infra/online_eval_provider.py (custom resource Lambda)
- Remove SKIP_ONLINE_EVAL logic from cdk-stack.ts and bin/cdk.ts
- Online eval is now purely declarative via agentcore.json onlineEvalConfigs[]
- To disable: set onlineEvalConfigs to [] in agentcore.json
Standards-Consulted: std.cdk.prefer-declarative
Standards-Gaps: none
Standards-Proposed: none
* fix: align model IDs with available Bedrock models
Update all model ID references to use the exact identifiers from
`aws bedrock list-foundation-models`:
- AGENT_MODEL_ID: us.anthropic.claude-sonnet-4-6 (not -20250929-v1:0)
- FAST_MODEL_ID: us.anthropic.claude-3-5-haiku-20241022-v1:0
- JUDGE_MODEL_ID: us.anthropic.claude-sonnet-4-6
The previous IDs (with -20250929-v1:0 suffix) were invalid and caused
runtime ValidationException on agent invocation.
Standards-Consulted: std.config.model-id-consistency, std.bedrock.verify-model-id-format
Standards-Gaps: none
Standards-Proposed: std.config.model-id-consistency, std.bedrock.verify-model-id-format, std.bedrock.validate-model-before-deploy
* fix: add --target dev to all agentcore CLI commands
This project uses a named target 'dev' in aws-targets.json (not the
default target). All CLI invocations must explicitly pass --target dev.
Standards-Consulted: std.deploy.target-dev
Standards-Gaps: none
Standards-Proposed: std.deploy.target-dev
* docs: overhaul documentation to describe current state only
- Remove SKIP_ONLINE_EVAL references throughout
- Fix model IDs in all documentation
- Remove .kiro references from public docs
- Merge duplicate Online Eval sections in README
- Fix GUARDRAIL_ID default (auto-creates, not skips)
- Fix target name in troubleshooting (dev, not default)
- Add --target dev to all documented deploy commands
- Add Declarative vs Imperative section to ARCHITECTURE.md
- Remove stale development process docs (8 root-level .md files)
- Pad all README tables for aligned vertical bars
Standards-Consulted: std.docs.current-state-only, std.deploy.target-dev, std.config.model-id-consistency
Standards-Gaps: none
Standards-Proposed: std.docs.current-state-only
* chore: update agentcore config and CDK dependencies
- agentcore.json: add onlineEvalConfigs, policyEngines, gateway config
- aws-targets.json.template: minor format fix
- CDK packages: update @aws/agentcore-cdk dependency range
Standards-Consulted: std.cdk.prefer-declarative, std.deploy.target-dev
Standards-Gaps: none
Standards-Proposed: none
* feat: improve agent resilience and MCP client handling
- main.py: add graceful degradation when MCP tools unavailable,
safe fallback to LLM-only mode on tool initialization failure
- mcp_client/client.py: add get_all_mcp_clients_safe() with error
collection instead of hard failure
- trigger: minor fix
- show_ticket.sh: minor fix
Standards-Consulted: std.agentcore.mcp-sigv4-auth
Standards-Gaps: none
Standards-Proposed: none
* feat: add end-to-end test script
scripts/test-e2e.sh publishes a ticket to SNS, polls DynamoDB for
resolution (10s intervals, 120s timeout), and asserts status=Resolved
with a non-empty resolution_comment. Exits 0 on pass, 1 on fail.
Usage:
./scripts/test-e2e.sh # sample ticket
./scripts/test-e2e.sh /path/to/ticket.json # custom ticket
Standards-Consulted: std.deploy.target-dev
Standards-Gaps: none
Standards-Proposed: none
* feat: upgrade FAST_MODEL_ID to Claude Haiku 4.5
Replace claude-3-5-haiku-20241022-v1:0 (legacy, access-gated) with
claude-haiku-4-5-20251001-v1:0 (current, available in account).
Verified: model invocable, E2E test passes all 3 tiers (LOW/HIGH/CRITICAL).
Standards-Consulted: std.config.model-id-consistency, std.bedrock.verify-model-id-format, std.bedrock.validate-model-before-deploy
Standards-Gaps: none
Standards-Proposed: none
* fix: add .gitignore with lib/ un-ignore for CDK source files
The root .gitignore excludes lib/ globally (for compiled JS output in
other samples). This project's CDK TypeScript source lives in
agentcore/cdk/lib/ and must be tracked. Add project-level .gitignore
with negation rules so 'git add' works without -f flag.
Standards-Consulted: std.git.no-deployed-state
Standards-Gaps: none
Standards-Proposed: none
* docs: remove sample-level LICENSE, inherit from monorepo root
Remove the local MIT-0 LICENSE file from it-incident-response-agent.
The sample should inherit the root repo's Apache 2.0 license, consistent
with all other samples in 02-use-cases/.
* feat: migrate observability to declarative agentcore.json, add Cedar policy steering
- Move OTEL/X-Ray env vars to agentcore.json runtimes[].envVars[]
- Add instrumentation.enableOtel: true
- Add onlineEvalConfigs[] (4 built-in evaluators, 100% sampling)
- Add policyEngines[] with Cedar policies (resource is AgentCore::Gateway)
- Add policyEngineConfiguration to gateway (LOG_ONLY mode)
- Remove ~150 lines of imperative CDK (custom resource, env var overrides)
- L3 construct now handles gateway lambda:InvokeFunction automatically
- Rename target from 'default' to 'dev'
- Fix trigger Lambda runtimeSessionId length (min 33 chars)
- Add Cedar policy syntax steering file
- Update README with CLI commands for online-eval + policy-engine
* chore: remove .kiro/ from git tracking, add to .gitignore
* fix: update evaluate.py
* refactor: simplify evaluate.py to retrieve online eval results
Replace complex on-demand evaluation with a script that queries the
online evaluation results log group. The continuous online evaluation
(agentcore.json onlineEvalConfigs[]) scores all invocations automatically.
Standards-Consulted: std.docs.current-state-only
Standards-Gaps: none
Standards-Proposed: none
* chore: clean up developer-only artifacts and reduce consumer confusion
- Remove AGENTS.md from tracking (Kiro AI context, not for consumers)
- Reorganize .gitignore with categories, add developer-only file exclusions
- Rename docs/online-eval-workaround.md → online-evaluation.md
- Trim ARCHITECTURE.md auth deep-dive (link to authentication-guide.md)
- Simplify README: collapse manual path, clarify CLI-first section,
remove duplicate env-var table, consolidate Configure section
* feat: add OTEL span attributes, tool-call hooks, e2e test, and eval reporter
- Add ticket.id/priority/requester_id/mode as OTEL span attributes for
end-to-end trace correlation via CloudWatch Transaction Search
- Add Strands BeforeToolCallEvent/AfterToolCallEvent hooks for per-tool
call timing in runtime logs
- Add scripts/e2e_test.py with live status polling, log tailing, and
post-resolution tool call timeline
- Rewrite scripts/evaluate.py to parse online eval results with summary
(avg scores by evaluator) and detailed per-trace breakdown
* remove: deprecate 02-use-cases/it-incident-response-agent (v1)
The v1 sample at 02-use-cases/it-incident-response-agent/ is superseded by
02-use-cases/automation-agents/it-incident-response-agent/, which is an
evolved version of the same use case with significant improvements:
- CLI-first workflow (agentcore.json + agentcore deploy) vs raw CDK
- Zero external prerequisites to deploy (Auth0/Jira optional, not required)
- All 6 AgentCore services demonstrated (adds Policy Engine, Guardrails)
- Production patterns: DLQ, idempotency, cost routing, graceful degradation
- Local dev support (agentcore dev with hot-reload)
- L3 CDK constructs instead of L1 CfnResource boilerplate
- Comprehensive documentation and design-decisions ADRs
The v1 code used raw L1 CfnResource constructs with mandatory Auth0 +
Jira dependencies, making it inaccessible for quick-start consumers.
All v1 functionality (Jira integration, Auth0 CUSTOM_JWT, Atlassian 3LO,
online evaluation) is preserved in the automation-agents version as
optional toggles.
* style: fix lint and format issues for CI compliance
Python (ruff):
- Remove unused imports: get_all_mcp_clients, get_streamable_http_mcp_client
(main.py), os (jira_oauth_provider.py), time (seeder.py)
- Auto-format 8 files to pass ruff format check
TypeScript (Prettier):
- Auto-format cdk-stack.ts, infra-construct.ts, bin/cdk.ts
All CI checks now pass:
- ruff check: 0 errors
- ruff format --check: 21 files formatted
- prettier --check: all files pass
- tsc --noEmit: compiles clean
- agentcore validate: Valid
* chore: gitignore eval doc build artifacts
* Potential fix for pull request finding 'CodeQL / Clear-text logging of sensitive information'
Co-authored-by: Copilot Autofix powered by AI <62310815+github-advanced-security[bot]@users.noreply.github.com>
Signed-off-by: Will Matos <wilmatos@amazon.com>
* Potential fix for pull request finding 'CodeQL / Clear-text logging of sensitive information'
Co-authored-by: Copilot Autofix powered by AI <62310815+github-advanced-security[bot]@users.noreply.github.com>
Signed-off-by: Will Matos <wilmatos@amazon.com>
* fix: resolve ruff lint errors in e2e_test.py
- F541: Remove extraneous f-prefix on string without placeholders (line 157)
- F841: Remove unused variable base_ts (line 270)
Standards-Consulted: std.python.ruff-lint-clean
Standards-Gaps: none
Standards-Proposed: none
* fix: additional e2e_test.py improvements
* fix: e2e_test.py refinements
* fix: code review improvements - error handling, threading, cleanup
- Wrap _resolve_ticket DDB call in try/except (non-fatal failure)
- Add DEBUG log for silent OTEL ImportError
- Remove unused imports in mcp_client/jira.py
- Add docstrings to memory/session.py
- Move import threading to top-level in e2e_test.py
- Replace mutable list stop flag with threading.Event
Verified: all Python files parse cleanly, tsc --noEmit passes,
agentcore validate passes, CDK tests pass, 50-ticket E2E test 100% pass.
* feat: enable Transaction Search observability and tighten IAM scoping
- Add OTEL/GenAI observability env vars (AGENT_OBSERVABILITY_ENABLED,
message-content capture, application signals) to agentcore.json
- Add Transaction Search custom resource Lambda to route X-Ray segments
to CloudWatch Logs (aws/spans) for online eval ingestion
- Scope trigger Lambda to bedrock-agentcore:InvokeAgentRuntime on the
specific runtime ARN (least privilege)
- Add resource-based lambda:InvokeFunction permissions for Gateway targets
- Make _fail_ticket DDB write non-fatal with exception logging
- Misc lint/cleanup in lambdas and e2e_test.py
* docs: reflect auto-enabled Transaction Search via custom resource
Transaction Search is now provisioned automatically by the CDK stack
(transaction_search.py custom resource) when onlineEvalConfigs is set,
so manual 'aws application-signals start-monitoring' is no longer required.
Update README, online-evaluation.md, and ARCHITECTURE.md accordingly.
* refactor: remove redundant L3-provided config, restore Memory resource
- Remove redundant XRayTracing IAM statement from RuntimeAdditionalPolicy (L3 RuntimeExecutionRole already grants xray put-trace perms); keep CloudWatch Logs Insights
- Remove instrumentation.enableOtel from agentcore.json (OTEL wrapping provided by Dockerfile CMD opentelemetry-instrument for Container build)
- Move AGENT_MODEL_ID/FAST_MODEL_ID into agentcore.json runtimes[].envVars[] (declarative, was imperative addPropertyOverride)
- Restore Memory resource (ITIncidentAgentMemory, SUMMARIZATION, namespace incidents/{actorId}) so memory code is backed by a provisioned resource instead of a no-op
- Enable ACTIVE X-Ray tracing on trigger Lambda for full service-map coverage
- Update README/ARCHITECTURE/online-evaluation docs; add 'agentcore add memory' CLI instructions
* docs: fix stale references in auth and schema docs
- authentication-guide: JIRA_MCP_URL is a hardcoded constant set by CDK, not auto-derived from JIRA_SITE_URL
- custom-jwt-auth-upgrade: correct stale function name _get_oauth_token() -> _create_custom_jwt_client()
- .llm-context/README: remove reference to non-existent mcp.ts schema file
* fix: deploy blockers for Memory namespace and gateway target IAM ordering
- Memory SUMMARIZATION namespace requires {sessionId}: incidents/{actorId} -> incidents/{actorId}/{sessionId} (CreateMemory validation failed without it). Retrieval still works via the incidents/{actorId} prefix (prefix matching).
- GatewayTargets now explicitly depend on the gateway role DefaultPolicy. The AgentCoreMcp L3 creates the lambda:InvokeFunction policy but does not order targets after it, causing deterministic 'Gateway execution role lacks permission' CREATE_FAILED. Replaces the ineffective resource-based fn.addPermission approach.
- README: Getting Started section (CI doc gate), Memory namespace + sessionId note
- ARCHITECTURE: corrected Memory namespace note
* style: lint fixes (ruff/pylint) with no behavior change
- main.py: remove unused imports GATEWAY_URL, MEMORY_ID (ruff F401); they are imported where actually used in mcp_client/memory modules
- mcp_client/client.py: remove unnecessary else-after-return (pylint R1705)
- mcp_client/client.py, jira.py: scoped 'pylint: disable=missing-kwoa' with justification on @requires_access_token-decorated calls (decorator injects access_token; pylint cannot see the transform)
* style: black formatting in lambda handlers (no behavior change)
- transaction_search.py: collapse wrapped log line
- ticket_event_handler.py: expand ALLOWED_FIELDS set to multi-line, collapse RuntimeError
* fix: auto-trigger KB ingestion on deploy by passing DataSourceId to seeder
The seeder gates ingestion on 'kb_id and data_source_id', but the CDK only passed KnowledgeBaseId — DataSourceId was never wired through, so start_ingestion_job never ran and the auto-created KB stayed empty until manual ingestion.
Capture the data source id on InfraConstruct (knowledgeBaseDataSourceId) and pass DataSourceId to the TriggerSeeder custom resource. README updated: ingestion is now automatic; manual command kept as optional re-index.
* fix: validate ticket_id/issue_key in agent entrypoint to avoid uncaught KeyError
payload['ticket_id'] was read outside the main try block, so a malformed/direct invoke missing ticket_id (and issue_key) raised an uncaught KeyError that crashed the invoke generator with no structured response. Use payload.get() with an explicit guard that yields a structured Failed result and returns; tighten is_jira_mode to bool(JIRA_MCP_URL).
* fix: agent entrypoint robustness
Guard model output extraction; empty output raises ValueError so the ticket is marked Failed (requires human processing) rather than resolved with a fallback. Sanitize both title and description through the guardrail. Remove dual logger alias. Avoid unbound agent reference when both initializations fail in prompt mode.
Findings #3, #6, #7, #8
* fix: memory cross-session retrieval
Use namespace_path (prefix match) instead of namespace (exact match) so prior-session memories are retrieved during enrichment.
Finding #4
* fix: atomic change-request write + reason field
Use transact_write_items for the dual DynamoDB write so the change request and audit records commit atomically. Add the reason parameter to the tool schema and wire it through, making the RequireReasonForChangeRequest Cedar policy functional.
Findings #5, #12
* fix: DynamoDB Decimal serialization in tools
Convert DynamoDB Decimal values to native int/float before json.dumps so tool responses serialize correctly.
Finding #15
* fix: trigger region + KB ingestion re-run + QueryKb env
Pass region_name to boto3 clients in the ticket event trigger. Add commonEnv to QueryKbFn. Bump the seeder Version to 3 so the KB data source ingestion re-runs.
Findings #9, #10, #14
* fix: e2e span duration unit
Correct the span duration unit conversion in the e2e test.
Finding #11
* style: pylint cleanups in e2e_test and evaluate scripts
- Drop unused loop variable, specify UTF-8 encoding on open()
- Simplify elif-after-return in _score_bar
- Add main() docstrings and wrap long lines
* style: wrap long lines in agent and lambda modules (C0301)
* style: apply ruff format (line-length 120)
* chore: rename folder
* feat(docs): update docs
* chore(usecases) - restructure under workflow
* feat(it-incident-agent): stream real-time pipeline stages + evaluations UI + demo
- Agent (main.py): emit real SSE stage events at each pipeline phase
(guardrail, memory, tools, diagnose, per-tool-call, persist, emit)
using agent.stream_async() instead of blocking agent() call
- Docs: add Real-Time Progress Streaming section to ARCHITECTURE.md
- README: add demo GIF showcasing the full workflow
- lambdas/tools: minor cleanup in create_change_request.py
- .pylintrc: project lint config
* style: add blank line before _stage_event function
* chore: update .gitignore for AgentCore CLI and CDK artifacts
* feat: add AgentCore CDK infrastructure (DynamoDB, Lambda, Cognito, EventBridge)
* feat: implement dual-agent claims processor with memory and MCP Gateway
* feat: update Lambda tool handlers with input validation and routing
* feat: add one-command deploy/destroy scripts and .env.example
* feat: add E2E test suite, Cedar tests, lint script, and unit tests
* docs: add architecture, deployment guide, ADRs, and update README
* refactor: remove legacy infra/ directory (replaced by agentcore/cdk/)
* chore: add gitignore negations for claims-agent CDK lib and Dockerfile
* fix: switch test_invoke.py to SigV4 auth (Runtime uses AWS_IAM, not JWT)
* docs: regenerate architecture diagrams and fix Runtime auth description
* style: fix import sorting and trailing whitespace (ruff auto-fix)
* chore: add wilmatos to CONTRIBUTORS.md
* style(event-driven-claims-agent): apply ruff format
* chore(event-driven-claims-agent): improve lint.sh with verbose output, format check, and tsc
* fix(claims-agent): use SigV4 for Runtime invocation, not JWT
The Runtime uses IAM (SigV4) auth; CUSTOM_JWT is only for the Gateway.
- Trigger Lambda: replace Cognito JWT flow with SigV4 signing
- test_e2e.py: switch from Bearer token to SigV4, relax assertions
to match actual agent streaming output format
- Dockerfile: add non-root user and healthcheck (CKV_DOCKER_2/3)
- test_local.py: add URL scheme validation, suppress S310 lint
---------
Signed-off-by: Will Matos <wilmatos@amazon.com>
Signed-off-by: Akarsha Sehwag <akshseh@amazon.de>
Co-authored-by: Akarsha Sehwag <akshseh@amazon.de>
Co-authored-by: Akarsha Sehwag <akarsha15010@iiitd.ac.in>
Co-authored-by: Will Matos <wmatosjr@users.noreply.github.com>
Co-authored-by: Copilot Autofix powered by AI <62310815+github-advanced-security[bot]@users.noreply.github.com>
* Harness GA Samples — AWS Skills, S3 Filesystem, Step Functions, Builder Agent (#1688)
* Add Harness GA samples: AWS Skills, S3 filesystem, Step Functions, builder agent
Four new samples under 01-features/01-harness for the Harness GA launch:
- 01-advanced-examples/08-aws-skills: enable native AWS Skills from the AWS
Agent Toolkit via the awsSkills source (all / glob / specific / mixed modes)
- 01-advanced-examples/09-s3-filesystem: mount an S3 Files access point as the
agent filesystem; includes s3_knowledge_base.py — a persistent LLM-wiki
knowledge base (ingest / query / lint) on the S3 mount
- 01-advanced-examples/10-stepfunctions-orchestration: a Step Functions STANDARD
state machine that drives the harness lifecycle (create -> poll READY ->
invoke -> delete) via a Lambda task worker
- 02-use-cases/03-aws-builder-agent: harness + AWS Skills compose an AWS
engineering agent that designs and scaffolds a serverless app
All samples use the GA boto3 SDK shapes (create_harness/invoke_harness) with no
internal endpoints or configuration. Updated the harness README index.
* Rename s3_knowledge_base.py to s3_llm_wiki.py to avoid Bedrock Knowledge Bases confusion
The LLM-wiki sample is a self-maintained markdown wiki on the agent's
filesystem, unrelated to the Amazon Bedrock Knowledge Bases feature. Rename the
file and scrub 'knowledge base' wording (mount default /mnt/wiki, pages/ subdir)
across the script and READMEs to prevent customer confusion.
* Fix S3 filesystem + Step Functions samples after end-to-end testing
Verified the samples against the live GA API and fixed real bugs found:
S3 filesystem (s3_filesystem.py, s3_llm_wiki.py):
- S3 Files mounts REQUIRE VPC network mode — add networkConfiguration
(networkMode=VPC + subnets + security groups) to create_harness; add
required --subnet-ids/--security-group-ids CLI args.
- Execution role needs s3files permissions (ListMountTargets/Get*/ClientMount/
ClientWrite/ClientRootAccess), not plain s3:* — correct the attached policy.
- README: document VPC + mount-target prerequisites and the new args.
Step Functions (stepfunctions_orchestration.py):
- Lambda role also needs *AgentRuntime* actions (CreateHarness provisions an
underlying AgentRuntime) — without them CreateHarness returns AccessDenied.
- CheckStatus Choice now matches CREATE_FAILED/UPDATE_FAILED so a failed
creation fails fast instead of looping on the WaitForReady default forever.
* Rework Step Functions sample to native integration; fix S3 Files IAM/VPC guidance
Step Functions: replace the Lambda task-worker approach with the native
arn:aws:states:::bedrockagentcore:invokeHarness service integration — a single
Task state, no Lambda, no glue. The state machine's own role gets
bedrock-agentcore:InvokeHarness; output is Converse-shaped
(Output.Message.Content[].Text). Rewrote the README accordingly.
S3 filesystem: scope the execution-role IAM to what the runtime actually needs —
s3files:GetAccessPoint (unscoped, validated at harness create time) plus
s3files:ClientMount/ClientWrite (conditioned on the access point). Document that
S3 Files mounts require VPC network mode with private subnets that have egress
(route to a NAT gateway); public subnets do not connect. Same IAM fix applied to
the s3_llm_wiki.py sample.
* Remove Step Functions sample; renumber to avoid collision with merged samples
Upstream main now occupies advanced-examples 01-12 (incl. 06-async-step-function,
08-gemini-model-provider, 09-openai-model-provider). Renumber our two samples to
free slots and drop the Step Functions sample (upstream added its own):
- delete 10-stepfunctions-orchestration
- 08-aws-skills -> 13-aws-skills
- 09-s3-filesystem -> 14-s3-filesystem
Updated all cross-references (harness index README, builder-agent note).
* docs: fix evaluations next step link (#1679)
* Add Weather Agent use case with harness, gateway, guardrails, evaluation and observability (#1648)
* Add Weather Agent use case with harness, gateway, guardrails, evaluations, and observability
* fix: resolve lint errors and security finding (unused imports, f-strings, bind to localhost)
* fix: address PR feedback (README clarifications, region detection, graceful eval error, trace
display)
* feat: add Skills (xlsx report generation) and Optimization (system prompt recommendations)
* feat: add Skills (xlsx report generation) and Optimization (system prompt recommendations)
* fix: address PR feedback (README clarifications, region detection, graceful eval error, trace
display)
* Agentcore optimization nys (#1722)
* Add failure insights sample (insights.py) and update README
- Add insights.py: runs FailureAnalysis, UserIntent, and ExecutionSummary
batch insight jobs on the HR Assistant agent. Supports --generate-traces
to send curated failure-mode sessions, --online to create a recurring
daily OnlineEvaluationConfig, and --insight to select individual insight
types. Uses both aws/spans and the runtime log group as data sources.
- Update README with a full Failure Insights section covering all three
insight types, data source requirements, CLI examples, and how to chain
insights into a system prompt recommendation.
* sample for agentcore insights feature with SDK scripts and CLI examples
* agent loops image
* move CLI insights step to after deploy and baseline eval in optimization workflow
* remove step 0 from CLI examples
* add insights.py description in How It Works section
* adding docs links to readme
* rename failure insights to insights throughout
* fix pylint and ruff issues in insights.py
- Add encoding="utf-8" to file read/write calls
- Wrap long lines to stay within 100-char limit
- Add pylint disable comments for intentional broad-exception-caught
- Rename loop variable to avoid module-scope naming false positive
- Remove f-string prefix from string literals without placeholders (ruff fix)
* remove generated state and config files
* remove TEST_LOG.md from registry
* scope aws-targets gitignore to exclude gateway config files
* clearing outputs from NBs
* fix ruff/pylint line-length conflict in optimize folder
* update repository structure in readme
* updating main readme
* replace account numbers and resource IDs with placeholders in optimization notebook
* Amazon Bedrock AgentCore Gateway websearch samples and usecases (#1721)
* Add AgentCore Web Search Tool samples and workshop content
- 01-features: new 03-web-search folder with setup, raw MCP, Strands,
and LangChain samples; updated requirements.txt with pinned versions
- 06-workshops: new 03-Agent-Core-web-search workshop with 6 notebooks
covering gateway setup, Strands agent, LangChain agent, CVE scanner,
earnings brief, and iterative research pattern
- 06-workshops/05-AgentCore-tools/README.md: added Web Search Tool section
* feat: add deep-research-agent with auto-provisioning and search privacy notices
- Add deep-research-agent use case with iterative Plan/Search/Reflect/Synthesize loop
- Add gateway_setup.py with auto-detect/prompt/provision flow (no hard prerequisites)
- Add search privacy callout to all 03-web-search README files
- Fix model ID default to use cross-region inference profile
- Add user-friendly error handling for auth failures
* Update finance budget assistant for Claude Haiku 4.5 (#1742)
- Update BUDGET_SYSTEM_PROMPT in lab1: the previous prompt broke the
structured output instructions with Claude Haiku 4.5
- Remove outdated note about enabling model access for Claude 3.7 Sonnet
- Update references from Anthropic Claude 3.7 Sonnet to Claude Haiku 4.5
in lab1 and README
* Add skip-extraction sample and rename 08-redrive to 08-manage-extraction (#1746)
Adds a runnable sample demonstrating extractionMode=SKIP on CreateEvent,
which stores events in short-term memory without triggering long-term
extraction. Renames the folder to 08-manage-extraction to cover both
skip and redrive as extraction lifecycle controls.
Co-authored-by: Gal Goldman <galgold@amazon.com>
* feat(memory): add multi-region-replication example (#1747)
* feat(memory): add multi-region-replication example
* feat(memory): update func
* chore: ruff formatting
* feat(use-cases): add multi-ISV orchestration sample (Salesforce + SAP) (#1640)
* feat(use-cases): add multi-ISV orchestration sample — Salesforce + SAP
Add a standalone use case under 02-use-cases/multi-isv-orchestration/
demonstrating how to connect Salesforce Lightning Platform and AWS for
SAP MCP Server to a single AgentCore Gateway, enabling cross-system AI
agent workflows through one unified MCP endpoint.
Three Jupyter notebooks walk through:
- 01: Salesforce as integration target (CustomOauth2, 43 tools)
- 02: SAP MCP Server as MCP target (9 tools, read-only default)
- 03: Cross-ISV queries (Customer 360, pipeline reconciliation)
Includes gateway_mcp_client.py utility, architecture diagrams, and
documented workarounds (Content-Type, domainName, org hibernation).
Originally proposed under 01-tutorials/02-AgentCore-gateway/ in
#1487; relocated to 02-use-cases/ to fit the new repo structure
where 01-features/ is CLI-only and end-to-end samples live under
02-use-cases/.
* security(multi-isv): scope Gateway IAM role and harden notebook inputs
Address threat-model review feedback on the multi-ISV orchestration sample:
- Scope the Gateway execution-role policy: split the single Resource:"*"
statement into four scoped statements (bedrock:InvokeModel limited to the
Claude model/inference-profile ARNs, secretsmanager:GetSecretValue limited
to the bedrock-agentcore-* secret prefix, iam:PassRole limited to the role
itself with a PassedToService condition); remove the unused s3:GetObject.
- Stop printing the Cognito client secret in the NB01 summary; show the
describe-user-pool-client retrieval command instead.
- Validate SF_DOMAIN against a strict [A-Za-z0-9-]+ pattern to prevent
redirecting tool calls to an attacker-controlled host.
- Update the README Disclaimer to reflect the scoped IAM policy.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* docs(multi-isv): add sample to workflow-automation category README
List multi-isv-orchestration in the 02-workflow-automation-agents
samples table after relocating the folder into that category.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* `feat(04-coding-agents): add sample 05 — autonomous coding agent with durable orchestration` (#1725)
* feat: add sample 05 — autonomous coding agent with durable orchestration
Event-driven headless coding backend on AgentCore Runtime with:
- Lambda Durable Function orchestrator (zero-cost suspension)
- 4 specialized runtimes (coding agent, sandbox, Swift sandbox, evaluator)
- Cedar policy enforcement at sandbox layer
- AgentCore Memory for cross-ticket learning
- Evaluator agent for read-only code review
- CDK deployment (8 stacks)
* docs: note that sandbox examples cover Python and Swift but are extendable to other frameworks
* fix: resolve ruff lint violations (E741, E401, F541) across four files
- Rename ambiguous variable l to lesson in list comprehensions
(shared/memory.py, orchestrator/handler.py)
- Split multi-import into separate statements (sandbox/app.py)
- Remove unnecessary f-string prefix (cdk/stacks/storage_stack.py)
- Add property-based tests verifying lint compliance and behavior preservation
* fix: suppress 8 false-positive ASH security findings with inline annotations
- Add # nosec B108 to intentional /tmp usage in isolated containers/microVMs
- Add # nosec B602 to sandboxed subprocess executor (sandbox/app.py)
- Add # nosec B108 to test files (assertions and fixtures, not real /tmp usage)
- Add #checkov:skip=CKV_DOCKER_3 to Dockerfile.swift (entrypoint.sh handles su)
- Each annotation includes justification for audit trail
* chore: remove AgentCore diagram icon PNGs from claims agent docs
* feat: refactor agent core — inline MCP client, add routing module, remove legacy modules
- Rewrite main.py: inline @requires_access_token MCP client builder,
structured output tools, confidence-based routing via routing.py
- Add routing.py: extracted pure routing logic (resolve_decision,
resolve_routing, decide_action) with safe fallback defaults
- Update config.py: L3 construct auto-generated env var support,
memory retrieval tuning params, credential provider config
- Update memory/session.py: configurable retrieval top_k and relevance
- Update Dockerfile: base image, uv sync steps
- Update pyproject.toml/requirements.txt/uv.lock: dependency refresh
- Remove mcp_client/ module: replaced by inline client in main.py
- Remove model/ module: load_model() now inline in main.py
- Remove parsing.py: regex parsing replaced by structured output tools
* feat: update CDK infra and agentcore.json for Identity-managed auth
- agentcore.json: add Cognito OAuth credential provider, OnlineEval
evaluator config, SEMANTIC + SUMMARIZATION memory strategies
- cdk-stack.ts: patch Lambda ARNs into gateway targets, wire CUSTOM_JWT
authorizer from Cognito discovery URL, add gateway target dependency
ordering fix, suppress mis-parsed target outputs
- infra-construct.ts: add DynamoDB GSIs (status, claim_id), S3 bucket
with EventBridge notifications, SNS topic, 6 Lambda tool functions +
trigger, EventBridge rule for S3 PutObject
- cdk.ts: updated stack instantiation with spec/mcpSpec/credentials
- package.json/package-lock.json: CDK dependency updates
* feat: update Lambda handlers and trigger for simplified routing
- trigger/handler.py: SigV4-signed HTTPS invocation to Runtime,
email parsing with claimant_email extraction, SSE response buffering
- list_pending_claims/handler.py: use GSI query instead of table scan
- notification/handler.py: updated response format, error handling
- resolve_claim/handler.py: write resolution to both Claims + Reviews tables
* feat: add deploy, teardown, auth, and observability scripts
* test: add unit tests for routing, structured output, Lambda handlers, trigger
* docs: update architecture, decisions, deployment, and configuration docs
- AGENTS.md: updated AI assistant context for current architecture
- README.md: updated quick start, project structure, commands
- .env.example: updated template for Identity-managed auth
- .gitignore: remove docs/decisions/ exclusion (publish ADRs)
- docs/ARCHITECTURE.md: updated system design, component descriptions
- docs/CONFIGURATION.md: updated config reference
- docs/README.md: updated index and prerequisites
- docs/deployment.md: updated deploy/verify/teardown steps
- docs/tutorial.md: updated guided walkthrough
- docs/decisions/0004: rewritten as Hybrid Auth (SigV4 + Cognito M2M)
- docs/decisions/0010: rewritten + renamed (Identity token vault)
- docs/decisions/0011-0012: now tracked (externalized config, GSI)
- docs/decisions/README.md: updated index with correct titles
* claims-agent: fire-and-forget trigger, resilient prompts, e2e timing
- Trigger Lambda: fire-and-forget invocation (read first 5 lines only,
don't buffer full SSE stream). Timeout bumped to 65s for cold starts.
- Trigger Lambda timeout in CDK: 60s → 90s to cover Runtime cold start.
- Processor prompt: graceful handling when lookup_policy fails, guard
against tool hallucination and fabricated policy details.
- E2E tests: add per-test timing, increase async wait to 90s.
- ARCHITECTURE.md: document fire-and-forget pattern.
- Add scripts/analyze_traces.py for OTEL trace analysis.
* feat(claims-agent): add cost-based model routing for Validation Agent
Route the Validation Agent (Phase 2) to Haiku via FAST_MODEL_ID,
saving ~80% cost and 3-8s latency per invocation. The validator
performs classification only (no tool use), making it an ideal
candidate for a smaller model.
Changes:
- Add FAST_MODEL_ID env var to config.py, agentcore.json, .env.example
- Update load_model() to accept fast=True for cost routing
- Wire get_validator() to use the fast model
- Add ADR-0013 documenting the decision
- Update AGENTS.md, ARCHITECTURE.md, CONFIGURATION.md
* claims-agent: cost routing (FAST_MODEL_ID) + deterministic Phase 3
Performance optimizations:
- Add FAST_MODEL_ID (Haiku) for Phase 2 Validation Agent — classification
task doesn't need Sonnet. Saves 3-8s and ~80% cost per invocation.
- Replace Phase 3 LLM call with direct MCPClient.call_tool_async() calls.
Routing is deterministic after Phase 2; no LLM needed. Saves 6-16s.
- Combined: pipeline goes from 3 Sonnet calls to 1 Sonnet + 1 Haiku + 0.
Measured: 65s (deployed old) → 28s (local new) = 56% reduction.
Code changes:
- config.py: add FAST_MODEL_ID env var
- main.py: load_model(fast=True) for validator, _call_tool() helper for
direct MCP calls, deterministic Phase 3 with non-fatal error handling
- agentcore.json: add FAST_MODEL_ID to runtime envVars
- .env.example: document FAST_MODEL_ID
Documentation:
- AGENTS.md: updated description, architecture diagram, invariant #8
- ARCHITECTURE.md: Phase 3 deterministic section, model routing line
- CONFIGURATION.md: FAST_MODEL_ID in both config tables
- deployment.md: fix --no-browser → --logs, port 3000 → 8080, Phase 3 output
- README.md: updated Phase 3 expected output
- ADR-0013: cost-based model routing decision
- ADR-0014: deterministic Phase 3 decision
- Skill ref: mcp-client-patterns.md (correct call_tool_async API)
* chore(claims-agent): sanitize agentcore.json and improve E2E tests/docs
- Replace env-specific Cognito discovery URL with PLACEHOLDER_USER_POOL_ID
- Add gateway exceptionLevel: NONE
- test_e2e.py: evidence-based assertions + --verbose flag
- deployment.md: add troubleshooting and config reference
- Add fix_credential_region.sh helper
- gitignore local backup and perf/design artifacts
* chore(claims-agent): placeholder the region in agentcore.json discovery URLs
Replace hardcoded us-west-2 with PLACEHOLDER_REGION in both the credentials
and gateway CUSTOM_JWT discovery URLs. Both values are overridden at deploy
(agentcore add credential / CDK authorizer patch), so this keeps the committed
config free of environment-specific region info.
* docs(claims-agent): refresh architecture diagram, fix ADRs, drop demo video
- Regenerate architecture.png (landscape high-level diagram) + commit
reproducible source docs/diagrams/architecture.py and AgentCore icons
- Reference architecture.png from README.md and docs/ARCHITECTURE.md
- Remove stale/duplicate ADR 0010-unsafe-unwrap.md (superseded by
0010-identity-token-vault-for-secrets.md; pattern not in code)
- Remove 6MB demo.mp4; README now embeds demo.gif (to be added)
* docs(claims-agent): add demo.gif and embed it in README
* fix(claims-agent): remove f-string prefix from strings without placeholders (F541)
* fix: apply ruff format to 5 Python files (CI lint fix)
---------
Signed-off-by: Will Matos <wilmatos@amazon.com>
Signed-off-by: Akarsha Sehwag <akshseh@amazon.de>
Co-authored-by: Akarsha Sehwag <akshseh@amazon.de>
Co-authored-by: Akarsha Sehwag <akarsha15010@iiitd.ac.in>
Co-authored-by: Will Matos <wmatosjr@users.noreply.github.com>
Co-authored-by: Copilot Autofix powered by AI <62310815+github-advanced-security[bot]@users.noreply.github.com>
Co-authored-by: Ed Fraga <39071108+edfragas@users.noreply.github.com>
Co-authored-by: 吴杨帆 <39647285+wyf027@users.noreply.github.com>
Co-authored-by: JobRamos <33988720+JobRamos@users.noreply.github.com>
Co-authored-by: Bharathi Srinivasan <bhrsrini@amazon.com>
Co-authored-by: Naga Gaddamu <zigeesha@hotmail.com>
Co-authored-by: philgut-aws <philgut@amazon.de>
Co-authored-by: gel-work <171434940+gel-work@users.noreply.github.com>
Co-authored-by: Gal Goldman <galgold@amazon.com>
Co-authored-by: Joachim Aumann <aumannjoachim@gmail.com>
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Co-authored-by: Sascha Möllering <smoell@amazon.de>
|
||
|
sumanayanamandra
|
bde5786ae2 |
Add AgentCore Web Search Tool managed connector for customer service agent (#1749)
* Add AgentCore Web Search Tool managed connector, remove third-party search dependency * added testing examples, verify no stale refs, update interceptor allowed tools, link to docs for IAM action * address review: remove tavily completely, delete WEB_SEARCH_README, fix checkov findings (PITR, DLQ, concurrency, scoped IAM) --------- Co-authored-by: Sumana Yanamandra <syanaman@amazon.com> |
||
|
Senthil
|
1229a4c385 |
feat: add AgentCore Gateway tool search plugin sample (#1622)
* feat: add AgentCore Gateway tool search plugin sample Add sample demonstrating semantic tool discovery with AgentCore Gateway and Strands Agents using travel-domain Lambda tools. * docs: add name to CONTRIBUTORS.md * refactor: convert notebook to separate Python scripts with detailed README - Replace notebook with deploy.py, invoke.py, cleanup.py, config.py - Add detailed README with architecture diagram and usage instructions - Add requirements.txt and .gitignore - Add architecture diagram from AgentCore SDK * fix: resolve lint errors in invoke.py and travel_tools.py - Remove f-prefix from strings without placeholders (F541) - Remove unused json import (F401) - Remove unused local variables interests and category (F841) * Adding benchmark results * feat: add intent relevance benchmark and update scaling benchmark --------- Signed-off-by: Senthil <34865595+senthilkumarmohan@users.noreply.github.com> |
||
|
Akarsha Sehwag
|
eb4b9609c4 |
feat(Memory): update namespaces to clarify multi-tenancy usecase (#1754)
* Update namespace docu Signed-off-by: Akarsha Sehwag <akshseh@amazon.de> * Update namespaces and memory strategies to show multi-tenancy ex Signed-off-by: Akarsha Sehwag <akshseh@amazon.de> --------- Signed-off-by: Akarsha Sehwag <akshseh@amazon.de> |
||
|
Diego Brasil
|
131cf1f980 |
Add Receipts IDP agent (Intelligent Document Processing) with a model degradation ladder (#1748)
* Add Receipts IDP agent (Intelligent Document Processing) with a model degradation ladder * style: apply ruff format (line-length 120) to receipts IDP sample Fixes the python-lint CI job on PR #1748. The CI runs both `ruff check` and `ruff format --check` against the repo-root config (line-length = 120); the sample previously only ran `ruff check`, so the formatter job failed on 26 files. Reformats those files (whitespace/wrapping only, no logic change) and updates the sample Makefile's `make lint` to run BOTH ruff commands at line-length 120, matching CI exactly. Adds `make format` to auto-fix. * fix(destroy): handle DELETE_FAILED / orphaned AgentCore control-plane resources Addresses PR review (destroy.sh). AgentCore control-plane resources (Runtime, Gateway, GatewayTarget, PolicyEngine, Evaluator) can fail to delete on the first pass due to control-plane resource ordering, leaving the stack in DELETE_FAILED. The old script's blind 'cloudformation wait' then exited with an opaque error and no guidance. destroy.sh now: 1. polls delete-stack to a terminal status; 2. on DELETE_FAILED, retries the plain delete once (transient ordering orphans usually clear); 3. if still stuck, re-issues delete with --retain-resources <ids> (valid only in DELETE_FAILED) so the stack + everything billable is removed; 4. prints each retained resource (ResourceType -> PhysicalResourceId) and points to the new docs/deployment.md recovery section for the exact bedrock-agentcore-control delete-* calls. docs/deployment.md gains a 'Teardown & DELETE_FAILED recovery' section with the verified delete-gateway / delete-gateway-target / delete-agent-runtime commands. All four teardown paths (clean, absent, recover, terminal) verified offline with a mocked aws CLI. * ci: scope workflow permissions (Checkov CKV2_GHA_1) Addresses PR review (ci.yml). Add top-level 'permissions: read-all' and per-job overrides so the workflow token is least-privilege: - unit: contents:read (checkout only) - e2e: contents:read + id-token:write (OIDC role assumption via aws-actions/configure-aws-credentials) Verified with checkov 3.3.6: CKV2_GHA_1 FAILED -> PASSED, no new finding. Token-scope audit confirms every step that uses a GITHUB_TOKEN scope still has it granted. --------- Co-authored-by: di-brasil <di-brasil@users.noreply.github.com> |
||
|
satveerkhurpa
|
426f41ce69 |
Add OBO training: overview, protocol reference, and end-to-end examples (#1753)
* Add OBO training: overview, protocol reference, and end-to-end examples * Address CodeQL + ruff findings ... * Apply ruff format across all Python files * Silence prints inside functions handling client_secret CodeQL's clear-text-logging query flags any print inside a function that receives client_secret as a parameter, even when the message body is a static string. Removed per-app progress prints from provider-creation functions (all four 01_create_providers.py + both 00_create_*_apps.py); callers still print a summary count after the loop. .env writes are unchanged — secrets must persist to file for the runtime. * Reformat obo-training with line-length=120 (matches root pyproject.toml) --------- Co-authored-by: Satveer Khurpa <khurpas@amazon.com> |
||
|
Sascha Möllering
|
73b5488bb1 |
feat(04-coding-agents): add sample 05 — autonomous coding agent with durable orchestration (#1725)
* feat: add sample 05 — autonomous coding agent with durable orchestration Event-driven headless coding backend on AgentCore Runtime with: - Lambda Durable Function orchestrator (zero-cost suspension) - 4 specialized runtimes (coding agent, sandbox, Swift sandbox, evaluator) - Cedar policy enforcement at sandbox layer - AgentCore Memory for cross-ticket learning - Evaluator agent for read-only code review - CDK deployment (8 stacks) * docs: note that sandbox examples cover Python and Swift but are extendable to other frameworks * fix: resolve ruff lint violations (E741, E401, F541) across four files - Rename ambiguous variable l to lesson in list comprehensions (shared/memory.py, orchestrator/handler.py) - Split multi-import into separate statements (sandbox/app.py) - Remove unnecessary f-string prefix (cdk/stacks/storage_stack.py) - Add property-based tests verifying lint compliance and behavior preservation * fix: suppress 8 false-positive ASH security findings with inline annotations - Add # nosec B108 to intentional /tmp usage in isolated containers/microVMs - Add # nosec B602 to sandboxed subprocess executor (sandbox/app.py) - Add # nosec B108 to test files (assertions and fixtures, not real /tmp usage) - Add #checkov:skip=CKV_DOCKER_3 to Dockerfile.swift (entrypoint.sh handles su) - Each annotation includes justification for audit trail |
||
|
Joachim Aumann
|
065f087a15 |
feat(use-cases): add multi-ISV orchestration sample (Salesforce + SAP) (#1640)
* feat(use-cases): add multi-ISV orchestration sample — Salesforce + SAP Add a standalone use case under 02-use-cases/multi-isv-orchestration/ demonstrating how to connect Salesforce Lightning Platform and AWS for SAP MCP Server to a single AgentCore Gateway, enabling cross-system AI agent workflows through one unified MCP endpoint. Three Jupyter notebooks walk through: - 01: Salesforce as integration target (CustomOauth2, 43 tools) - 02: SAP MCP Server as MCP target (9 tools, read-only default) - 03: Cross-ISV queries (Customer 360, pipeline reconciliation) Includes gateway_mcp_client.py utility, architecture diagrams, and documented workarounds (Content-Type, domainName, org hibernation). Originally proposed under 01-tutorials/02-AgentCore-gateway/ in #1487; relocated to 02-use-cases/ to fit the new repo structure where 01-features/ is CLI-only and end-to-end samples live under 02-use-cases/. * security(multi-isv): scope Gateway IAM role and harden notebook inputs Address threat-model review feedback on the multi-ISV orchestration sample: - Scope the Gateway execution-role policy: split the single Resource:"*" statement into four scoped statements (bedrock:InvokeModel limited to the Claude model/inference-profile ARNs, secretsmanager:GetSecretValue limited to the bedrock-agentcore-* secret prefix, iam:PassRole limited to the role itself with a PassedToService condition); remove the unused s3:GetObject. - Stop printing the Cognito client secret in the NB01 summary; show the describe-user-pool-client retrieval command instead. - Validate SF_DOMAIN against a strict [A-Za-z0-9-]+ pattern to prevent redirecting tool calls to an attacker-controlled host. - Update the README Disclaimer to reflect the scoped IAM policy. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * docs(multi-isv): add sample to workflow-automation category README List multi-isv-orchestration in the 02-workflow-automation-agents samples table after relocating the folder into that category. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com> |
||
|
Akarsha Sehwag
|
dea7328db0 |
feat(memory): add multi-region-replication example (#1747)
* feat(memory): add multi-region-replication example * feat(memory): update func * chore: ruff formatting |
||
|
gel-work
|
fb14ec6b3c |
Add skip-extraction sample and rename 08-redrive to 08-manage-extraction (#1746)
Adds a runnable sample demonstrating extractionMode=SKIP on CreateEvent, which stores events in short-term memory without triggering long-term extraction. Renames the folder to 08-manage-extraction to cover both skip and redrive as extraction lifecycle controls. Co-authored-by: Gal Goldman <galgold@amazon.com> |
||
|
philgut-aws
|
e74dccaacd |
Update finance budget assistant for Claude Haiku 4.5 (#1742)
- Update BUDGET_SYSTEM_PROMPT in lab1: the previous prompt broke the structured output instructions with Claude Haiku 4.5 - Remove outdated note about enabling model access for Claude 3.7 Sonnet - Update references from Anthropic Claude 3.7 Sonnet to Claude Haiku 4.5 in lab1 and README |
||
|
Naga Gaddamu
|
16df45472a |
Amazon Bedrock AgentCore Gateway websearch samples and usecases (#1721)
* Add AgentCore Web Search Tool samples and workshop content - 01-features: new 03-web-search folder with setup, raw MCP, Strands, and LangChain samples; updated requirements.txt with pinned versions - 06-workshops: new 03-Agent-Core-web-search workshop with 6 notebooks covering gateway setup, Strands agent, LangChain agent, CVE scanner, earnings brief, and iterative research pattern - 06-workshops/05-AgentCore-tools/README.md: added Web Search Tool section * feat: add deep-research-agent with auto-provisioning and search privacy notices - Add deep-research-agent use case with iterative Plan/Search/Reflect/Synthesize loop - Add gateway_setup.py with auto-detect/prompt/provision flow (no hard prerequisites) - Add search privacy callout to all 03-web-search README files - Fix model ID default to use cross-region inference profile - Add user-friendly error handling for auth failures |
||
|
Bharathi Srinivasan
|
a444af49f4 |
Agentcore optimization nys (#1722)
* Add failure insights sample (insights.py) and update README - Add insights.py: runs FailureAnalysis, UserIntent, and ExecutionSummary batch insight jobs on the HR Assistant agent. Supports --generate-traces to send curated failure-mode sessions, --online to create a recurring daily OnlineEvaluationConfig, and --insight to select individual insight types. Uses both aws/spans and the runtime log group as data sources. - Update README with a full Failure Insights section covering all three insight types, data source requirements, CLI examples, and how to chain insights into a system prompt recommendation. * sample for agentcore insights feature with SDK scripts and CLI examples * agent loops image * move CLI insights step to after deploy and baseline eval in optimization workflow * remove step 0 from CLI examples * add insights.py description in How It Works section * adding docs links to readme * rename failure insights to insights throughout * fix pylint and ruff issues in insights.py - Add encoding="utf-8" to file read/write calls - Wrap long lines to stay within 100-char limit - Add pylint disable comments for intentional broad-exception-caught - Rename loop variable to avoid module-scope naming false positive - Remove f-string prefix from string literals without placeholders (ruff fix) * remove generated state and config files * remove TEST_LOG.md from registry * scope aws-targets gitignore to exclude gateway config files * clearing outputs from NBs * fix ruff/pylint line-length conflict in optimize folder * update repository structure in readme * updating main readme * replace account numbers and resource IDs with placeholders in optimization notebook |
||
|
JobRamos
|
e720768134 |
Add Weather Agent use case with harness, gateway, guardrails, evaluation and observability (#1648)
* Add Weather Agent use case with harness, gateway, guardrails, evaluations, and observability * fix: resolve lint errors and security finding (unused imports, f-strings, bind to localhost) * fix: address PR feedback (README clarifications, region detection, graceful eval error, trace display) * feat: add Skills (xlsx report generation) and Optimization (system prompt recommendations) * feat: add Skills (xlsx report generation) and Optimization (system prompt recommendations) * fix: address PR feedback (README clarifications, region detection, graceful eval error, trace display) |
||
|
吴杨帆
|
4002ce6bef | docs: fix evaluations next step link (#1679) | ||
|
Ed Fraga
|
bb763db66c |
Harness GA Samples — AWS Skills, S3 Filesystem, Step Functions, Builder Agent (#1688)
* Add Harness GA samples: AWS Skills, S3 filesystem, Step Functions, builder agent Four new samples under 01-features/01-harness for the Harness GA launch: - 01-advanced-examples/08-aws-skills: enable native AWS Skills from the AWS Agent Toolkit via the awsSkills source (all / glob / specific / mixed modes) - 01-advanced-examples/09-s3-filesystem: mount an S3 Files access point as the agent filesystem; includes s3_knowledge_base.py — a persistent LLM-wiki knowledge base (ingest / query / lint) on the S3 mount - 01-advanced-examples/10-stepfunctions-orchestration: a Step Functions STANDARD state machine that drives the harness lifecycle (create -> poll READY -> invoke -> delete) via a Lambda task worker - 02-use-cases/03-aws-builder-agent: harness + AWS Skills compose an AWS engineering agent that designs and scaffolds a serverless app All samples use the GA boto3 SDK shapes (create_harness/invoke_harness) with no internal endpoints or configuration. Updated the harness README index. * Rename s3_knowledge_base.py to s3_llm_wiki.py to avoid Bedrock Knowledge Bases confusion The LLM-wiki sample is a self-maintained markdown wiki on the agent's filesystem, unrelated to the Amazon Bedrock Knowledge Bases feature. Rename the file and scrub 'knowledge base' wording (mount default /mnt/wiki, pages/ subdir) across the script and READMEs to prevent customer confusion. * Fix S3 filesystem + Step Functions samples after end-to-end testing Verified the samples against the live GA API and fixed real bugs found: S3 filesystem (s3_filesystem.py, s3_llm_wiki.py): - S3 Files mounts REQUIRE VPC network mode — add networkConfiguration (networkMode=VPC + subnets + security groups) to create_harness; add required --subnet-ids/--security-group-ids CLI args. - Execution role needs s3files permissions (ListMountTargets/Get*/ClientMount/ ClientWrite/ClientRootAccess), not plain s3:* — correct the attached policy. - README: document VPC + mount-target prerequisites and the new args. Step Functions (stepfunctions_orchestration.py): - Lambda role also needs *AgentRuntime* actions (CreateHarness provisions an underlying AgentRuntime) — without them CreateHarness returns AccessDenied. - CheckStatus Choice now matches CREATE_FAILED/UPDATE_FAILED so a failed creation fails fast instead of looping on the WaitForReady default forever. * Rework Step Functions sample to native integration; fix S3 Files IAM/VPC guidance Step Functions: replace the Lambda task-worker approach with the native arn:aws:states:::bedrockagentcore:invokeHarness service integration — a single Task state, no Lambda, no glue. The state machine's own role gets bedrock-agentcore:InvokeHarness; output is Converse-shaped (Output.Message.Content[].Text). Rewrote the README accordingly. S3 filesystem: scope the execution-role IAM to what the runtime actually needs — s3files:GetAccessPoint (unscoped, validated at harness create time) plus s3files:ClientMount/ClientWrite (conditioned on the access point). Document that S3 Files mounts require VPC network mode with private subnets that have egress (route to a NAT gateway); public subnets do not connect. Same IAM fix applied to the s3_llm_wiki.py sample. * Remove Step Functions sample; renumber to avoid collision with merged samples Upstream main now occupies advanced-examples 01-12 (incl. 06-async-step-function, 08-gemini-model-provider, 09-openai-model-provider). Renumber our two samples to free slots and drop the Step Functions sample (upstream added its own): - delete 10-stepfunctions-orchestration - 08-aws-skills -> 13-aws-skills - 09-s3-filesystem -> 14-s3-filesystem Updated all cross-references (harness index README, builder-agent note). |
||
|
Will Matos
|
97e89df7b6 |
feat(it-incident-response-agent): production-ready IT incident response agent with streaming and observability (#1724)
* feat(usecases): it-incident-response
* add jira integration
* fix minor memory usage points
* docs: add IT incident response agent project docs and assets
* config: add AgentCore project configuration and schema context
* infra: add CDK project for AgentCore L3 construct deployment
* feat: add Strands agent application (runtime, memory, MCP client, model)
* feat: add Lambda functions for tools, infra providers, and ticket trigger
* feat: add tool schemas, seed data, and knowledge base runbooks
* chore: add deployment, evaluation, and ticket utility scripts
* docs: comply with use-case README template (add details table and disclaimer)
* chore: allow esbuild install script in CDK project
* fix: create S3 Vectors bucket+index for KB (CFN does not auto-create)
The Bedrock KnowledgeBase CloudFormation resource with type S3_VECTORS
requires a pre-existing vector bucket and index. Passing an empty
s3VectorsConfiguration fails schema validation. The console's 'quick
create' auto-provisions these, but CloudFormation does not.
Changes:
- Import aws-cdk-lib/aws-s3vectors
- Create CfnVectorBucket (named per account+region)
- Create CfnIndex (float32, 1024 dims, cosine, no metadata keys)
- Wire IndexArn, IndexName, VectorBucketArn into KB storageConfiguration
- Grant KB role s3vectors:* actions on the vector index
- Add removalPolicy to bucket+index
- Add troubleshooting row to README
- Deleted the ROLLBACK_COMPLETE stack to unblock next deploy
* fix: resolve deploy issues (template envsubst, target name, IndexName removal)
- aws-targets.json.template: fix envsubst-incompatible default syntax
(${AWS_REGION:-us-west-2} → ${AWS_REGION}), change name 'dev' → 'default'
- scripts/deploy.sh: export AWS_REGION with default before envsubst runs
- infra-construct.ts: remove IndexName from s3VectorsConfiguration (caused
CFN 'oneOf 2 subschemas matched' validation error — only IndexArn +
VectorBucketArn are needed for BYO S3 Vectors)
- deployed-state.json: updated by successful deploy
* docs: update Quickstart to lead with deploy.sh as primary path
* chore: stop tracking deployed-state.json (deployment-specific, not shared)
* refactor(auth): rename OAUTH_PROVIDER_NAME to GATEWAY_OAUTH_PROVIDER_NAME
Scope auth env vars per boundary for clarity:
- OAUTH_PROVIDER_NAME -> GATEWAY_OAUTH_PROVIDER_NAME
- GATEWAY_AUDIENCE -> GATEWAY_OAUTH_AUDIENCE
Code maintains backward-compat fallback to legacy names.
CDK injects new names into Runtime env vars at deploy.
.env.example and agentcore/.env.local.example updated
with boundary labels (Boundary 2, Boundary 3).
* docs(auth): add authentication boundary guide
New docs/authentication-guide.md explains the 3 auth boundaries:
1. Runtime Inbound (SigV4 vs CUSTOM_JWT)
2. Gateway Outbound (AWS_IAM vs CUSTOM_JWT M2M)
3. Jira Outbound (USER_FEDERATION 3LO)
Covers: conceptual overview, env var reference, local dev
implications, troubleshooting, and why all 3 patterns exist.
* docs: fix local dev ports, CLI commands, and env var references
README.md:
- Add Port Mapping section (8081=Web UI, 8082=Runtime container)
- Add troubleshooting entry for workload access token error
- Fix agentcore invoke --dev (invalid) -> agentcore dev prompt
- Update config table to use GATEWAY_OAUTH_* var names
docs/custom-jwt-auth-upgrade.md:
- Update all OAUTH_PROVIDER_NAME -> GATEWAY_OAUTH_PROVIDER_NAME
- Update all GATEWAY_AUDIENCE -> GATEWAY_OAUTH_AUDIENCE
docs/ARCHITECTURE.md:
- Wrap ASCII diagrams in <details> tags for readability
* Fix S3 Vectors Knowledge Base schema: remove indexName from s3VectorsConfiguration
CloudFormation's AWS::Bedrock::KnowledgeBase type with S3_VECTORS storage
has a schema constraint (oneOf) that rejects when both indexArn and indexName
are provided. Pass only vectorBucketArn + indexArn to satisfy the schema.
The indexName is implicit in the indexArn and Bedrock manages metadata
internally during ingestion, so it's not needed in the configuration.
Fixes: Properties validation failed - 'only 1 subschema matches out of 2'
Verified: Stack now deploys successfully with CREATE_COMPLETE status
* feat: migrate online eval to declarative agentcore.json
Remove the custom resource workaround for Online Evaluation. The
AgentCoreOnlineEvaluationConfig L3 construct (@aws/agentcore-cdk
v0.1.0-alpha.34+) now handles dependency ordering automatically.
Changes:
- Delete lambdas/infra/online_eval_provider.py (custom resource Lambda)
- Remove SKIP_ONLINE_EVAL logic from cdk-stack.ts and bin/cdk.ts
- Online eval is now purely declarative via agentcore.json onlineEvalConfigs[]
- To disable: set onlineEvalConfigs to [] in agentcore.json
Standards-Consulted: std.cdk.prefer-declarative
Standards-Gaps: none
Standards-Proposed: none
* fix: align model IDs with available Bedrock models
Update all model ID references to use the exact identifiers from
`aws bedrock list-foundation-models`:
- AGENT_MODEL_ID: us.anthropic.claude-sonnet-4-6 (not -20250929-v1:0)
- FAST_MODEL_ID: us.anthropic.claude-3-5-haiku-20241022-v1:0
- JUDGE_MODEL_ID: us.anthropic.claude-sonnet-4-6
The previous IDs (with -20250929-v1:0 suffix) were invalid and caused
runtime ValidationException on agent invocation.
Standards-Consulted: std.config.model-id-consistency, std.bedrock.verify-model-id-format
Standards-Gaps: none
Standards-Proposed: std.config.model-id-consistency, std.bedrock.verify-model-id-format, std.bedrock.validate-model-before-deploy
* fix: add --target dev to all agentcore CLI commands
This project uses a named target 'dev' in aws-targets.json (not the
default target). All CLI invocations must explicitly pass --target dev.
Standards-Consulted: std.deploy.target-dev
Standards-Gaps: none
Standards-Proposed: std.deploy.target-dev
* docs: overhaul documentation to describe current state only
- Remove SKIP_ONLINE_EVAL references throughout
- Fix model IDs in all documentation
- Remove .kiro references from public docs
- Merge duplicate Online Eval sections in README
- Fix GUARDRAIL_ID default (auto-creates, not skips)
- Fix target name in troubleshooting (dev, not default)
- Add --target dev to all documented deploy commands
- Add Declarative vs Imperative section to ARCHITECTURE.md
- Remove stale development process docs (8 root-level .md files)
- Pad all README tables for aligned vertical bars
Standards-Consulted: std.docs.current-state-only, std.deploy.target-dev, std.config.model-id-consistency
Standards-Gaps: none
Standards-Proposed: std.docs.current-state-only
* chore: update agentcore config and CDK dependencies
- agentcore.json: add onlineEvalConfigs, policyEngines, gateway config
- aws-targets.json.template: minor format fix
- CDK packages: update @aws/agentcore-cdk dependency range
Standards-Consulted: std.cdk.prefer-declarative, std.deploy.target-dev
Standards-Gaps: none
Standards-Proposed: none
* feat: improve agent resilience and MCP client handling
- main.py: add graceful degradation when MCP tools unavailable,
safe fallback to LLM-only mode on tool initialization failure
- mcp_client/client.py: add get_all_mcp_clients_safe() with error
collection instead of hard failure
- trigger: minor fix
- show_ticket.sh: minor fix
Standards-Consulted: std.agentcore.mcp-sigv4-auth
Standards-Gaps: none
Standards-Proposed: none
* feat: add end-to-end test script
scripts/test-e2e.sh publishes a ticket to SNS, polls DynamoDB for
resolution (10s intervals, 120s timeout), and asserts status=Resolved
with a non-empty resolution_comment. Exits 0 on pass, 1 on fail.
Usage:
./scripts/test-e2e.sh # sample ticket
./scripts/test-e2e.sh /path/to/ticket.json # custom ticket
Standards-Consulted: std.deploy.target-dev
Standards-Gaps: none
Standards-Proposed: none
* feat: upgrade FAST_MODEL_ID to Claude Haiku 4.5
Replace claude-3-5-haiku-20241022-v1:0 (legacy, access-gated) with
claude-haiku-4-5-20251001-v1:0 (current, available in account).
Verified: model invocable, E2E test passes all 3 tiers (LOW/HIGH/CRITICAL).
Standards-Consulted: std.config.model-id-consistency, std.bedrock.verify-model-id-format, std.bedrock.validate-model-before-deploy
Standards-Gaps: none
Standards-Proposed: none
* fix: add .gitignore with lib/ un-ignore for CDK source files
The root .gitignore excludes lib/ globally (for compiled JS output in
other samples). This project's CDK TypeScript source lives in
agentcore/cdk/lib/ and must be tracked. Add project-level .gitignore
with negation rules so 'git add' works without -f flag.
Standards-Consulted: std.git.no-deployed-state
Standards-Gaps: none
Standards-Proposed: none
* docs: remove sample-level LICENSE, inherit from monorepo root
Remove the local MIT-0 LICENSE file from it-incident-response-agent.
The sample should inherit the root repo's Apache 2.0 license, consistent
with all other samples in 02-use-cases/.
* feat: migrate observability to declarative agentcore.json, add Cedar policy steering
- Move OTEL/X-Ray env vars to agentcore.json runtimes[].envVars[]
- Add instrumentation.enableOtel: true
- Add onlineEvalConfigs[] (4 built-in evaluators, 100% sampling)
- Add policyEngines[] with Cedar policies (resource is AgentCore::Gateway)
- Add policyEngineConfiguration to gateway (LOG_ONLY mode)
- Remove ~150 lines of imperative CDK (custom resource, env var overrides)
- L3 construct now handles gateway lambda:InvokeFunction automatically
- Rename target from 'default' to 'dev'
- Fix trigger Lambda runtimeSessionId length (min 33 chars)
- Add Cedar policy syntax steering file
- Update README with CLI commands for online-eval + policy-engine
* chore: remove .kiro/ from git tracking, add to .gitignore
* fix: update evaluate.py
* refactor: simplify evaluate.py to retrieve online eval results
Replace complex on-demand evaluation with a script that queries the
online evaluation results log group. The continuous online evaluation
(agentcore.json onlineEvalConfigs[]) scores all invocations automatically.
Standards-Consulted: std.docs.current-state-only
Standards-Gaps: none
Standards-Proposed: none
* chore: clean up developer-only artifacts and reduce consumer confusion
- Remove AGENTS.md from tracking (Kiro AI context, not for consumers)
- Reorganize .gitignore with categories, add developer-only file exclusions
- Rename docs/online-eval-workaround.md → online-evaluation.md
- Trim ARCHITECTURE.md auth deep-dive (link to authentication-guide.md)
- Simplify README: collapse manual path, clarify CLI-first section,
remove duplicate env-var table, consolidate Configure section
* feat: add OTEL span attributes, tool-call hooks, e2e test, and eval reporter
- Add ticket.id/priority/requester_id/mode as OTEL span attributes for
end-to-end trace correlation via CloudWatch Transaction Search
- Add Strands BeforeToolCallEvent/AfterToolCallEvent hooks for per-tool
call timing in runtime logs
- Add scripts/e2e_test.py with live status polling, log tailing, and
post-resolution tool call timeline
- Rewrite scripts/evaluate.py to parse online eval results with summary
(avg scores by evaluator) and detailed per-trace breakdown
* remove: deprecate 02-use-cases/it-incident-response-agent (v1)
The v1 sample at 02-use-cases/it-incident-response-agent/ is superseded by
02-use-cases/automation-agents/it-incident-response-agent/, which is an
evolved version of the same use case with significant improvements:
- CLI-first workflow (agentcore.json + agentcore deploy) vs raw CDK
- Zero external prerequisites to deploy (Auth0/Jira optional, not required)
- All 6 AgentCore services demonstrated (adds Policy Engine, Guardrails)
- Production patterns: DLQ, idempotency, cost routing, graceful degradation
- Local dev support (agentcore dev with hot-reload)
- L3 CDK constructs instead of L1 CfnResource boilerplate
- Comprehensive documentation and design-decisions ADRs
The v1 code used raw L1 CfnResource constructs with mandatory Auth0 +
Jira dependencies, making it inaccessible for quick-start consumers.
All v1 functionality (Jira integration, Auth0 CUSTOM_JWT, Atlassian 3LO,
online evaluation) is preserved in the automation-agents version as
optional toggles.
* style: fix lint and format issues for CI compliance
Python (ruff):
- Remove unused imports: get_all_mcp_clients, get_streamable_http_mcp_client
(main.py), os (jira_oauth_provider.py), time (seeder.py)
- Auto-format 8 files to pass ruff format check
TypeScript (Prettier):
- Auto-format cdk-stack.ts, infra-construct.ts, bin/cdk.ts
All CI checks now pass:
- ruff check: 0 errors
- ruff format --check: 21 files formatted
- prettier --check: all files pass
- tsc --noEmit: compiles clean
- agentcore validate: Valid
* chore: gitignore eval doc build artifacts
* Potential fix for pull request finding 'CodeQL / Clear-text logging of sensitive information'
Co-authored-by: Copilot Autofix powered by AI <62310815+github-advanced-security[bot]@users.noreply.github.com>
Signed-off-by: Will Matos <wilmatos@amazon.com>
* Potential fix for pull request finding 'CodeQL / Clear-text logging of sensitive information'
Co-authored-by: Copilot Autofix powered by AI <62310815+github-advanced-security[bot]@users.noreply.github.com>
Signed-off-by: Will Matos <wilmatos@amazon.com>
* fix: resolve ruff lint errors in e2e_test.py
- F541: Remove extraneous f-prefix on string without placeholders (line 157)
- F841: Remove unused variable base_ts (line 270)
Standards-Consulted: std.python.ruff-lint-clean
Standards-Gaps: none
Standards-Proposed: none
* fix: additional e2e_test.py improvements
* fix: e2e_test.py refinements
* fix: code review improvements - error handling, threading, cleanup
- Wrap _resolve_ticket DDB call in try/except (non-fatal failure)
- Add DEBUG log for silent OTEL ImportError
- Remove unused imports in mcp_client/jira.py
- Add docstrings to memory/session.py
- Move import threading to top-level in e2e_test.py
- Replace mutable list stop flag with threading.Event
Verified: all Python files parse cleanly, tsc --noEmit passes,
agentcore validate passes, CDK tests pass, 50-ticket E2E test 100% pass.
* feat: enable Transaction Search observability and tighten IAM scoping
- Add OTEL/GenAI observability env vars (AGENT_OBSERVABILITY_ENABLED,
message-content capture, application signals) to agentcore.json
- Add Transaction Search custom resource Lambda to route X-Ray segments
to CloudWatch Logs (aws/spans) for online eval ingestion
- Scope trigger Lambda to bedrock-agentcore:InvokeAgentRuntime on the
specific runtime ARN (least privilege)
- Add resource-based lambda:InvokeFunction permissions for Gateway targets
- Make _fail_ticket DDB write non-fatal with exception logging
- Misc lint/cleanup in lambdas and e2e_test.py
* docs: reflect auto-enabled Transaction Search via custom resource
Transaction Search is now provisioned automatically by the CDK stack
(transaction_search.py custom resource) when onlineEvalConfigs is set,
so manual 'aws application-signals start-monitoring' is no longer required.
Update README, online-evaluation.md, and ARCHITECTURE.md accordingly.
* refactor: remove redundant L3-provided config, restore Memory resource
- Remove redundant XRayTracing IAM statement from RuntimeAdditionalPolicy (L3 RuntimeExecutionRole already grants xray put-trace perms); keep CloudWatch Logs Insights
- Remove instrumentation.enableOtel from agentcore.json (OTEL wrapping provided by Dockerfile CMD opentelemetry-instrument for Container build)
- Move AGENT_MODEL_ID/FAST_MODEL_ID into agentcore.json runtimes[].envVars[] (declarative, was imperative addPropertyOverride)
- Restore Memory resource (ITIncidentAgentMemory, SUMMARIZATION, namespace incidents/{actorId}) so memory code is backed by a provisioned resource instead of a no-op
- Enable ACTIVE X-Ray tracing on trigger Lambda for full service-map coverage
- Update README/ARCHITECTURE/online-evaluation docs; add 'agentcore add memory' CLI instructions
* docs: fix stale references in auth and schema docs
- authentication-guide: JIRA_MCP_URL is a hardcoded constant set by CDK, not auto-derived from JIRA_SITE_URL
- custom-jwt-auth-upgrade: correct stale function name _get_oauth_token() -> _create_custom_jwt_client()
- .llm-context/README: remove reference to non-existent mcp.ts schema file
* fix: deploy blockers for Memory namespace and gateway target IAM ordering
- Memory SUMMARIZATION namespace requires {sessionId}: incidents/{actorId} -> incidents/{actorId}/{sessionId} (CreateMemory validation failed without it). Retrieval still works via the incidents/{actorId} prefix (prefix matching).
- GatewayTargets now explicitly depend on the gateway role DefaultPolicy. The AgentCoreMcp L3 creates the lambda:InvokeFunction policy but does not order targets after it, causing deterministic 'Gateway execution role lacks permission' CREATE_FAILED. Replaces the ineffective resource-based fn.addPermission approach.
- README: Getting Started section (CI doc gate), Memory namespace + sessionId note
- ARCHITECTURE: corrected Memory namespace note
* style: lint fixes (ruff/pylint) with no behavior change
- main.py: remove unused imports GATEWAY_URL, MEMORY_ID (ruff F401); they are imported where actually used in mcp_client/memory modules
- mcp_client/client.py: remove unnecessary else-after-return (pylint R1705)
- mcp_client/client.py, jira.py: scoped 'pylint: disable=missing-kwoa' with justification on @requires_access_token-decorated calls (decorator injects access_token; pylint cannot see the transform)
* style: black formatting in lambda handlers (no behavior change)
- transaction_search.py: collapse wrapped log line
- ticket_event_handler.py: expand ALLOWED_FIELDS set to multi-line, collapse RuntimeError
* fix: auto-trigger KB ingestion on deploy by passing DataSourceId to seeder
The seeder gates ingestion on 'kb_id and data_source_id', but the CDK only passed KnowledgeBaseId — DataSourceId was never wired through, so start_ingestion_job never ran and the auto-created KB stayed empty until manual ingestion.
Capture the data source id on InfraConstruct (knowledgeBaseDataSourceId) and pass DataSourceId to the TriggerSeeder custom resource. README updated: ingestion is now automatic; manual command kept as optional re-index.
* fix: validate ticket_id/issue_key in agent entrypoint to avoid uncaught KeyError
payload['ticket_id'] was read outside the main try block, so a malformed/direct invoke missing ticket_id (and issue_key) raised an uncaught KeyError that crashed the invoke generator with no structured response. Use payload.get() with an explicit guard that yields a structured Failed result and returns; tighten is_jira_mode to bool(JIRA_MCP_URL).
* fix: agent entrypoint robustness
Guard model output extraction; empty output raises ValueError so the ticket is marked Failed (requires human processing) rather than resolved with a fallback. Sanitize both title and description through the guardrail. Remove dual logger alias. Avoid unbound agent reference when both initializations fail in prompt mode.
Findings #3, #6, #7, #8
* fix: memory cross-session retrieval
Use namespace_path (prefix match) instead of namespace (exact match) so prior-session memories are retrieved during enrichment.
Finding #4
* fix: atomic change-request write + reason field
Use transact_write_items for the dual DynamoDB write so the change request and audit records commit atomically. Add the reason parameter to the tool schema and wire it through, making the RequireReasonForChangeRequest Cedar policy functional.
Findings #5, #12
* fix: DynamoDB Decimal serialization in tools
Convert DynamoDB Decimal values to native int/float before json.dumps so tool responses serialize correctly.
Finding #15
* fix: trigger region + KB ingestion re-run + QueryKb env
Pass region_name to boto3 clients in the ticket event trigger. Add commonEnv to QueryKbFn. Bump the seeder Version to 3 so the KB data source ingestion re-runs.
Findings #9, #10, #14
* fix: e2e span duration unit
Correct the span duration unit conversion in the e2e test.
Finding #11
* style: pylint cleanups in e2e_test and evaluate scripts
- Drop unused loop variable, specify UTF-8 encoding on open()
- Simplify elif-after-return in _score_bar
- Add main() docstrings and wrap long lines
* style: wrap long lines in agent and lambda modules (C0301)
* style: apply ruff format (line-length 120)
* chore: rename folder
* feat(docs): update docs
* chore(usecases) - restructure under workflow
* feat(it-incident-agent): stream real-time pipeline stages + evaluations UI + demo
- Agent (main.py): emit real SSE stage events at each pipeline phase
(guardrail, memory, tools, diagnose, per-tool-call, persist, emit)
using agent.stream_async() instead of blocking agent() call
- Docs: add Real-Time Progress Streaming section to ARCHITECTURE.md
- README: add demo GIF showcasing the full workflow
- lambdas/tools: minor cleanup in create_change_request.py
- .pylintrc: project lint config
* style: add blank line before _stage_event function
* chore: update .gitignore for AgentCore CLI and CDK artifacts
* feat: add AgentCore CDK infrastructure (DynamoDB, Lambda, Cognito, EventBridge)
* feat: implement dual-agent claims processor with memory and MCP Gateway
* feat: update Lambda tool handlers with input validation and routing
* feat: add one-command deploy/destroy scripts and .env.example
* feat: add E2E test suite, Cedar tests, lint script, and unit tests
* docs: add architecture, deployment guide, ADRs, and update README
* refactor: remove legacy infra/ directory (replaced by agentcore/cdk/)
* chore: add gitignore negations for claims-agent CDK lib and Dockerfile
* fix: switch test_invoke.py to SigV4 auth (Runtime uses AWS_IAM, not JWT)
* docs: regenerate architecture diagrams and fix Runtime auth description
* style: fix import sorting and trailing whitespace (ruff auto-fix)
* chore: add wilmatos to CONTRIBUTORS.md
* style(event-driven-claims-agent): apply ruff format
* chore(event-driven-claims-agent): improve lint.sh with verbose output, format check, and tsc
* fix(claims-agent): use SigV4 for Runtime invocation, not JWT
The Runtime uses IAM (SigV4) auth; CUSTOM_JWT is only for the Gateway.
- Trigger Lambda: replace Cognito JWT flow with SigV4 signing
- test_e2e.py: switch from Bearer token to SigV4, relax assertions
to match actual agent streaming output format
- Dockerfile: add non-root user and healthcheck (CKV_DOCKER_2/3)
- test_local.py: add URL scheme validation, suppress S310 lint
---------
Signed-off-by: Will Matos <wilmatos@amazon.com>
Signed-off-by: Akarsha Sehwag <akshseh@amazon.de>
Co-authored-by: Akarsha Sehwag <akshseh@amazon.de>
Co-authored-by: Akarsha Sehwag <akarsha15010@iiitd.ac.in>
Co-authored-by: Will Matos <wmatosjr@users.noreply.github.com>
Co-authored-by: Copilot Autofix powered by AI <62310815+github-advanced-security[bot]@users.noreply.github.com>
|