AI Agent Workflow Collection

Per-bead quality audit over the open backlog — flags empty/thin descriptions, label drift, stale YAGNIs, mis-prioritised nice-to-haves, obvious splits/epics, and duplicates. Judges description quality, not template conformance (template-section checks are opt-in). Read-only sweep that produces a proposal report; mutations apply only on explicit approval, and the dangerous ones (close, supersede, promote, split) are confirmed one at a time. Delegates splitting to /triage and cross-system linking to /tracking-sweep (Jira) or /trello-beads (Trello), calibrating to whichever tracker the project actually uses.

Check CircleCI build status and failed job logs for the current GitHub repository. Use when asked whether CircleCI is green, failing, pending, or when needing CircleCI job logs.

Browse handoff files saved by /wrap-up and pick one to resume. Lists this repo's handoffs in full (including ones whose worktree has been pruned) and summarises other repos by count. Companion to /wrap-up and /landscape.

Prune handoffs that no longer point at live work — superseded (a newer handoff continues the thread), done (PR merged, all beads closed, branch landed, or Jira ticket Done), or stale (branch gone / PR closed) — and archive them so the /handoffs picker stays focused. Read-only until you confirm; archives (never deletes).

Read-only audit of Pi routing and second-opinion consensus model IDs against the active Pi catalog and public live model metadata; reports when Pi or configured models merit review without editing config.

Derive a conventional Claude Code session name from the branch ticket, active bead, open PR, and what the session is actually doing — then print a paste-ready /rename line. Use when a session's auto-name is generic and you want it legible in the session list.

Ticket-scoped, read-only status dashboard that shows blockers, unmet requirements, verification evidence, working-copy state, tracking drift, and concrete untracked follow-ups.

Pre-merge gate — verify a PR is green, approved, in sync, and free of obvious risk, then (on explicit approval) squash-merge it. Composes /pr-status, /contract-check, and /review-pr rather than reimplementing them.

Deep release-readiness gate for a single letterbox service — checks CI green, contracts in sync, deploy-order prereqs satisfied, feature toggle present, and unpushed work vs the live deploy. Emits a gate table and a single verdict. Use before deciding to ship one service.

Reply to PR review comments after addressing them. Resolves conversations where changes were made. Uses polite tone for humans, terse factual responses for AI bots. IMPORTANT - Always use this skill (not raw gh api calls) when replying to PR comments, including after manually addressing review feedback.

Apply a lightweight common-sense YAGNI/KISS lens to find the smallest maintainable implementation before or during ordinary coding.

Sort, dedupe, and audit Claude settings.json / settings.local.json files at user and project level. Flags risky permissions, broken references, glob-subsumed entries, syntax errors, and cross-file duplicates that could be promoted up the hierarchy. Mechanical fixes auto-apply, judgment calls are presented as a triage list.

Full pre-PR quality gauntlet — runs clean-code, verify-task, code-review, pedantic-review, /review, /security-review, and tiered /second-opinion in increasing cost order. Halts on Must-Fix tier findings (failing tests, security, blocking-severity bugs), emits beads for the rest, iterates the heavy phases up to twice. Local cousin of /ultrareview.

After a push or merge, watch a CircleCI + FluxCD deploy until it lands — CircleCI green for the commit, then the k8s Deployment's image tag moves off its pre-push baseline and pods go ready — then run a read-only smoke test scoped to the change. Goal-terminating loop — stops when the rollout lands and the smoke completes, or when it fails.

Start a recurring release-gatekeeper loop — runs /release-manager on an adaptive cadence (fast when something is in flight, backing off when settled) until end of day in a dedicated tab. Prompts to push / defer / cancel as services become ready; runs attended.

After a merge, watch the GitHub Actions deploy run until the gating job lands, then run a smoke test scoped to the change (browser for UI, GET for read-only API) against staging. Goal-terminating loop — stops when the deploy lands and the smoke completes, or when it fails.