Tristan-Hauber Separate lenses, run any subset via lens=git|worklist|codex|claude|all (default all).
git: scripts/workflow_signals.py git <repo> — commit cadence, revert rate, PR turnaround (open->first-review, open->merge), stale-branch count. Read-only, no LLM calls in the script itself.
worklist: scripts/workflow_signals.py worklist <tasks.md path> — open-item count, oldest unresolved item's age, items with an unresolved blocker annotation.
codex: defer entirely to the weekly-codex-usage-review and weekly-skill-gap-review automations (outside this catalogue) — do not re-derive Codex usage or skill-gap findings here; summarise their latest memory only.
claude: best-effort from the current session only, unless mistake-analysis's shared store has Claude-sourced records — this lens has no persistent cross-session log to draw on and must say so rather than imply full coverage.
Merge lens outputs into one compact report: metric, current value, trend if known, one-line concern. For a concern worth acting on, emit a GO: <exact next step> or NOGO: <short reason> per finding, in the same style as the workflow-improvement pipeline's recommender stage (outside this catalogue) — never edit skills, rules, or the worklist directly.