phoenix-skills-audit

Run the audit against origin/main over a chosen window (default 7 days). It triages commits affecting Python clients, TypeScript clients, CLI, and APIs, then edits the phoenix-tracing, phoenix-cli, and phoenix-evals skill files in place with verified signatures and examples.

Ask for a skills audit or skill drift check. The workflow fetches origin/main, lists commits with git log, reads the actual diffs rather than trusting commit messages, and patches any skill snippet whose signature or import path no longer matches the tree.

Yes. The workflow pairs each Python change with its TypeScript mirror by grepping the equivalent package for the same symbol family. If a feature is genuinely Python-only or TS-only, that is flagged in the run summary as a finding.

Only phoenix-tracing, phoenix-cli, and phoenix-evals, including their SKILL.md files and references directories. Changes belonging to other skills like phoenix-server or phoenix-frontend are noted in the run summary but never edited.

Local main is routinely stale by dozens of commits, so auditing it silently misses everything shipped after the last git pull. The workflow always runs git fetch origin main first and records the audited SHA in the run summary for reproducibility.

Internal refactors with no public surface, dependency bumps, test-only or CI changes, frontend-only changes, release bookkeeping commits, and feature flags such as *_DANGEROUSLY_* or *_EXPERIMENTAL_* env vars are skipped and listed in the run summary for reviewer confirmation.