Recommended action: Check `date -u` against a trusted source and enable NTP; don't install custom CAs or disable verification. Option: Fix the system clock [evidence: official_recommended_action] Applies when: See trigger Steps: 1. date -u; compare with a reliable time source.
Cause (Documented platform behavior): With thinking enabled, the final assistant tool-use turn must begin with its original signed thinking/redacted_thinking block; clients that strip it violate the rule. Fix status: workaround_only Limitations: - Issue page read via WebFetch sum
Recommended action: Raise client_max_body_size in the relevant context (e.g. 50m) to exceed your largest legitimate payload, keeping app-level limits; for ingress-nginx use the proxy-body-size annotation (not verified here). Option: Raise the limit where the upload route is proxi
Recommended action: Keep at most one unnamed tool type and set server_label on all MCP tools; pass load_prompts=False (or equivalent), don't call send_ping(), always stream tools/call, or use the Agent Framework FoundryToolbox class; call tools as server_label.tool_name. Option:
Recommended action: Make bind address and probe URL identical (e.g. start server with --host 127.0.0.1 and use url http://127.0.0.1:PORT), raise webServer.timeout for build+serve commands, set stdout:'pipe' or DEBUG=pw:webserver to see server output, and ensure the ready URL retu
Recommended action: Check every NDJSON object for an 'error' key and for done: true; treat a stream that ends without done: true as failed. Option: Validate stream termination [evidence: official_recommended_action] Applies when: See trigger Steps: 1. for line in stream:
Recommended action: Verify at receipt time (before enqueueing), keep hosts NTP-synced, and for fixtures re-sign test payloads with a fresh timestamp instead of replaying captured ones. Option: Verify on receipt, not at processing time [evidence: official_recommended_action] Appli
Recommended action: Verify the connection's refresh URL; add offline_access to the scope when creating the OAuth connection; have the user re-consent if refresh tokens were revoked. Option: Enable refresh [evidence: official_recommended_action] Steps: 1. Recreate OAuth connection
Recommended action: pip install 'strands-agents[openai]'; set region; ensure the credential chain resolves; use endpoint='bedrock-runtime' for OpenAI GPT models served via cross-Region profile ids. Option: Install extra, set region, check credentials [evidence: official_recommend
Recommended action: Catch LengthFinishReasonError, inspect err.completion.usage, and retry with a larger token budget (account for reasoning tokens) or smaller output; handle content filter separately. Option: Retry with a larger budget [evidence: official_recommended_action] App
Recommended action: For Playwright MCP start the server with --ignore-https-errors (or env PLAYWRIGHT_MCP_IGNORE_HTTPS_ERRORS) or put ignoreHTTPSErrors in browser.contextOptions; in library code pass ignoreHTTPSErrors: true to browser.newContext()/launchPersistentContext() with t
Recommended action: session_quota_exceeded: stop sessions not needing compute, delete stale sessions, or request a limit increase. regional_session_quota_exceeded: retry with exponential backoff and jitter or use another region; open a support request if persistent. Option: Branc
Recommended action: Use a forwarding service (smee.io, gh webhook forward) for local testing; for private systems allowlist GitHub hook IPs from GET /meta and refresh periodically. Option: Forward or allowlist [evidence: official_recommended_action] Applies when: See trigger Step
Recommended action: Upgrade boto3/botocore to a release whose bedrock-runtime model includes the feature; pin minimum versions in requirements. Option: Upgrade boto3/botocore [evidence: official_recommended_action] Steps: 1. pip install -U boto3 botocore Expected: New blocks roun
Recommended action: Use project.get_openai_client() then openai.conversations.create(); update @azure/ai-projects to 2.4.0+; replace create_agent() with create_version()/CreateAgentVersionAsync()/createAgentVersion(). Option: Use the OpenAI client from the project [evidence: offi
Recommended action: Wrap unions in an object field (class Out(BaseModel): result: A | B), break $ref-to-$ref cycles, or disable strict for that schema. Option: Wrap the union in an object [evidence: documented_workaround] Applies when: See trigger Steps: 1. class Output(BaseModel
Recommended action: Hover the terminal tab to check shell integration quality (needs Rich); fix PROMPT_COMMAND to a single function or disable conflicting prompt plugins in VS Code terminals; use manual shell-integration install for complex setups; update VS Code/Copilot Chat. Op
Recommended action: Acknowledge immediately with 2xx after verifying the signature, enqueue the payload, process asynchronously; dedupe on X-GitHub-Delivery for redeliveries. Option: Ack fast, process async [evidence: official_recommended_action] Applies when: See trigger Steps:
Recommended action: List build workflows for the policy and delete failed/obsolete ones (list_automated_reasoning_policy_build_workflows + delete) before starting a new one; do not blind-retry. Option: Delete stale build workflows first [evidence: documented_workaround] Steps: 1.
Recommended action: Use a valid cpu/memory pair from Sandbox sizes; poll version status until active before invoking; stream :logstream to fix readiness/startup; append &force=true to DELETE to cascade-delete sessions; use the container deploy doc for image-based agents. Option: