Lean and token-efficient: every line carries non-obvious domain knowledge (fingerprint semantics, warehouse freshness lag, parquet scan pruning) with no padding or explanation of concepts Claude already knows.

Provides concrete, executable guidance — named MCP tools, a shape-reading table mapping readings to next steps, and copy-ready SQL pointed to in the referenced queries file — rather than abstract direction.

A clearly sequenced workflow (start wide → three shapes → trunk-break culprit → flaky corroboration) with explicit validation checkpoints (confidence check before naming a culprit, warehouse freshness check before trusting a boundary, verify the failure window matches).

The body is an overview that signals a single one-level-deep reference (references/investigation-queries.md, verified to exist) for the SQL detail; content is appropriately split and easy to navigate.

Lists concrete actions ('Investigates a specific CI failure to a verdict: whose fault, which commit, who wrote it, and whether it's fixed') and names the exact warehouse views and logs it works from — no vague language.

Explicitly answers both what (investigates a CI failure to a verdict) and when (a 'Use for …' clause listing concrete triggers), satisfying the explicit-trigger requirement.

Quotes natural phrases users actually say ('who broke master', 'why did this test fail in CI', 'is this failure my PR's fault or everyone's', 'is this test flaky or actually broken', 'when did this failure start').

A clear niche plus explicit negative triggers routing overlapping needs to sibling skills ('Not for aggregate CI health, cost, or merge bottlenecks (use diagnosing-ci-and-merge-bottlenecks) and not for building saved insights').