coworklisted
Install: claude install-skill tomcounsell/ai
# Cowork — Claude Code Routines
> **Pattern-design trigger addressed; deployment/live-verification still open.**
> This skill codifies a pattern that shipped once as a fully live, verified
> pilot (`docs/infra/cowork-sentry-triage.md`) and has now been exercised a
> second time at the **code** level: issue #2068's `pr-review-audit` migration
> proved the guard-shim generalizes to a structurally different audit (a
> `GH_REPO` project-synthesis guard, a fixed lookback window, and per-PR title
> dedup — see `docs/features/cowork-tasks.md`). The original "reviewable on
> first reuse" trigger is satisfied at the pattern-design level. What's still
> open: `pr-review-audit`'s CMA deployment and live-fire verification have not
> happened yet — see `docs/infra/cowork-pr-review-audit.md` for the pending
> operator steps. Treat the guidance below as reviewed-and-holding, not yet
> doubly-proven end-to-end in production.
## Repo Context Probe
If `.claude/skill-context/cowork.md` exists, read it and honor its declarations; otherwise use the generic defaults described below.
The context file is where a repo declares its local-scheduling-vs-routine decision rule, any existing local reflection/cron system a candidate task might be migrating away from, and repo-specific executables for checking what's currently scheduled locally or filing the routine's output. When the file is absent (the common case in a foreign repo), follow the generic baseline below.
## What a routine is
A **Claude