Skill: analyze-ci-failures

Analyze CI failures on an Azure SDK for .NET pull request and post a GitHub comment with actionable fix instructions.

When Invoked

Trigger phrases: "analyze CI failures", "fix CI", "why is CI failing", "help with CI", "analyze PR checks", "CI is red", "failing checks".

Inputs

The user must provide a PR number, PR URL, or pipeline build ID. If not provided, ask the user.

Workflow

1. Gather information

• Fetch PR details, check statuses, changed files, and workflow runs using GitHub MCP tools.
• Extract service directory and package name from changed file paths (sdk/<service>/<package>/).
• Identify the CI provider from each failed check URL before fetching logs:
  • For an Azure DevOps URL (dev.azure.com), extract the buildId. Use azure-sdk-mcp:azsdk_analyze_pipeline when available.
  • If that tool is unavailable, query the public Azure DevOps timeline API with curl: https://dev.azure.com/<organization>/<project>/_apis/build/builds/<buildId>/timeline?api-version=7.1. Inspect issues on failed task records first. If the issues are insufficient, fetch the failed record's log.url.
  • For a GitHub Actions URL, use github-mcp-server-get_job_logs with return_content: true.
• Do not use GitHub Actions job-log tools for Azure DevOps checks; their job IDs and run IDs are unrelated.

2. Identify failures

Classify each failure using the CI check mapping and log symptom patterns below. Also inspect the PR's code directly (e.g., read generated files for compile errors, check for missing scaffolding files).

Diagnosis must be based on the failed task's actual error text:

• Quote the decisive error and name the affected file or project.
• Do not infer the root cause from the check name, because Build/Pack jobs also run changelog verification, package validation, and ApiCompat.
• Do not call a failure transient or flaky unless the logs contain infrastructure evidence (for example, an agent disconnect, timeout, service outage, or network failure). A documentation-only last commit does not establish flakiness because packaging validates files such as CHANGELOG.md.
• If logs cannot be retrieved, state that the root cause is unconfirmed and provide the failed task link. Do not substitute a guessed compilation or ApiCompat diagnosis.

3. Post a comment

Compose a GitHub comment with:

• Header: ## 🔍 CI Failure Analysis for PR #<number>
• Summary table: All checks with ✅ ❌ ⏳ status
• Per-failure sections: Specific to THIS PR — include actual error messages, affected files, and concrete fix commands with <service>/<package> filled in
• Quick fix command at the end if applicable

Before posting, check existing comments for ## 🔍 CI Failure Analysis to avoid duplicates.

CI Check Name → Failure Mapping

These are the Azure DevOps and GitHub checks that run on SDK PRs. The check names are repo-specific and not discoverable from general knowledge.

Log Symptom → Root Cause Mapping

These are exact strings/patterns to search for in CI logs. They are specific to this repo's scripts and not inferrable from general knowledge.

New Package Checklist

For PRs that introduce a new SDK package (all files are added, no prior version exists), also check for these commonly missing scaffolding files:

• CHANGELOG.md
• README.md
• Directory.Build.props (must import parent props)
• ci.mgmt.yml (for management-plane) or CI pipeline config
• api/*.cs (public API surface listings — generated via Export-API.ps1)
• assets.json (test recording assets)

Without ci.mgmt.yml, the Azure DevOps CI pipeline won't trigger — this is a common reason for checkenforcer staying permanently pending.