pipeline-investigation

Pipeline Investigation

Investigate AzDO pipeline failures that aren't Helix test failures — build errors, infrastructure tooling crashes, validation test flakiness, and artifact cascade failures. Complements helix-investigation by covering everything else in the pipeline.

When to Use This Skill

• User has an AzDO build URL with a non-Helix failure (build step, validation, infra task)
• User asks "why did the pipeline fail" and the failure is in a build/scan/validation task
• User wants to know how often a specific failure occurs (frequency/trend analysis)
• User sees exit code null, 1ES PT errors, or MSBuild failures
• User wants to understand artifact cascade failures ("missing artifacts from prior build")
• User asks about SourcelinkTests, Binary Analysis Scan, or installer validation failures

Output Formats

This skill produces two distinct report types. Match the format to the request:

Health Assessment ("pipeline health", "are builds passing", "pipeline status")

Follow references/health-assessment-format.md. Output MUST include these two tables:

1. Failed Builds Table — every failed build, classified and investigated: | Build | Type | Source | Failure Detail |

   • Type: Rolling, Forward Flow, or Other PR (classify via gh pr view)
   • Source: Rolling → branch name. Forward Flow → target ← source-repo. Other PR → short description.

2. Summary Table — pass/fail breakdown by build type: | Type | Completed | ✅ Pass | ❌ Fail | Pass Rate |

3. Failure Trends Table (conditional — include when 3+ builds in scope and at least one pattern recurs; cap at top 5): | Pattern | Hits | Window | Status |

   • Status: ❌ No issue filed, ✅ Fix merged, 🔄 Known issue (link), ⏳ Fix in progress

See the reference for build classification rules, branch filtering, and codeflow analysis methodology.

Save the report: After presenting the health assessment, save it to reviews/pipeline-health-<slug>-YYYY-MM-DD-HHMM.md in the repo, where <slug> identifies the pipeline or scope (e.g., unified-build, runtime-ci). Include the timestamp to ensure uniqueness across multiple investigations per day. Include a Methodology section at the end documenting data sources and classification approach.