Recommended action: In Actions set lfs: true on actions/checkout (or run git lfs pull after checkout); elsewhere install git-lfs, run git lfs install, then git lfs pull. Detect pointers by the 'version https://git-lfs.github.com/spec/v1' first line. Option: Fetch LFS content expl
Recommended action: Check the HTML for 'App unavailable in region'; check network connectivity/proxy; use Homebrew cask or WinGet as alternative (same hosts though). Option: Diagnose the returned page, then use alternate installer [evidence: documented_workaround] Applies when: s
Recommended action: Assign a Microsoft 365 Copilot license to the user or enable pay-as-you-go; also confirm same-tenant access. Option: License the user [evidence: official_recommended_action] Steps: 1. Assign M365 Copilot license or enable pay-as-you-go Expected: Tool returns r
Recommended action: Pass model-specific params via model_extras on ChatCompletionsClient.complete (header added automatically) or upgrade to the addressed version; for Azure OpenAI deployments prefer the official openai client. Evidence basis (self-declared by the contributing ch
Recommended action: Prefer restructuring: run untrusted fork code in a plain pull_request workflow and only use pull_request_target/workflow_run for trusted steps without executing PR code. Only if you have reviewed the risks, set allow-unsafe-pr-checkout: true on that checkout s
Recommended action: Add PowerShell to PATH or install PowerShell 7, or set CLAUDE_CODE_GIT_BASH_PATH in settings.json env to ...\Git\bin\bash.exe (not git-bash.exe). Option: Point CLAUDE_CODE_GIT_BASH_PATH at bin\bash.exe [evidence: official_recommended_action] Applies when: see
Recommended action: Invoke the agent with a delegated user token (OBO); do not use app-only auth for SharePoint grounding. Option: Use delegated user identity [evidence: official_recommended_action] Steps: 1. Acquire a user token in the client app 2. Call the agent
Recommended action: Omit response_format for gpt-image-* models and read data[].b64_json; use output_format/background/quality for GPT image options. In LiteLLM set additional_drop_params: ['response_format'] (drop_params alone did not help per report). Option: Drop response_form
Recommended action: Confirm the remote is really the intended repo (wrong remote URL is a common cause). If joining two independent histories is intended, use --allow-unrelated-histories once; otherwise re-clone the remote and re-apply local changes. Option: Verify remote, then m
Recommended action: Run `aws sts get-caller-identity` with the same AWS_PROFILE; do `aws sso login` before starting; raise CLAUDE_CODE_AWS_CHAIN_RESOLVE_TIMEOUT_MS only for legitimately slow interactive helpers (aws-vault + MFA). Option: Pre-resolve credentials outside Claude Cod
Recommended action: Grant the project managed identity Container Registry Repository Reader on the ACR, ensure the registry's azureADAuthenticationAsArmPolicy is enabled, verify image URI/tag, and register the subscription provider. Option: Fix ACR access [evidence: official_reco
Recommended action: Use aspectRatio (e.g. '16:9') and the provider's image size option instead of size; assert on result.warnings in tests. Option: Map size to aspectRatio per provider [evidence: official_recommended_action] Steps: 1. Check warnings Expected: The call succeeds or
Recommended action: Fetch full history before pushing (git fetch --unshallow, or checkout with fetch-depth: 0), then push. Option: Unshallow before pushing [evidence: official_recommended_action] Applies when: Shallow CI checkouts Steps: 1. git fetch --unshallow origin (or action
Recommended action: Allow local listeners; upgrade to v2.1.268+ (falls back to OS-assigned port); for fixed port free it or register a different redirect URI and set MCP_OAUTH_CALLBACK_PORT. Fix: Upgrade and allow local listener [evidence: released_fix] Applies when: see problem
Recommended action: Treat as non-idempotent: have the model issue a new tool call (new call_id); make tools idempotent where possible. Option: Start a fresh tool call [evidence: documented_workaround] Steps: 1. Prompt/let the model issue a new call 2. Design tools
Recommended action: Pass a provider model instance, e.g. google.video('veo-...') or vertex.video('veo-...'), instead of a string id. Option: Import the provider [evidence: official_recommended_action] Steps: 1. Use provider.video('model-id') Expected: The call succeeds or the fai
Recommended action: Report to the server operator: the fix is server-side (return the advertised issuer). Temporary: MCP_SDK_GENERATION=v1 disables the check (reduces security). Option: Fix issuer on the authorization server [evidence: official_recommended_action] Applies when: s
Recommended action: Use function tools on realtime agents (wrap hosted capabilities behind your own function tool or a hosted MCP where supported), and keep hosted tools on text agents reached via handoff/agent-as-tool. Option: Replace hosted tools with function tools [evidence:
Recommended action: Express pace/style in natural language inside the text/prompt; for multi-speaker, put style cues in the speaker-labelled transcript; check warnings in tests. Option: Remove speed/language [evidence: official_recommended_action] Steps: 1. Put style directions i
Recommended action: Run /mcp and re-authenticate; for headersHelper fix the helper output; for static header update config; for insufficient_scope add the scope to pinned oauth.scopes before re-auth. Option: Re-authenticate via /mcp, adjusting pinned scopes if needed [evidence: o