Capabilities

Use when you notice you've written or applied a pattern that is generic, non-trivial, and likely to recur — extract it into a draft capability skill and stage it as a PR to agent-playbook so the next agent inherits it. Triggers on "this code is reusable", "every project does this", "I keep typing

never used

Use at session start (after reading project canon, before brainstorming or planning) to find cross-project knowledge that applies to the task at hand. Greps agent-playbook/canon/ for entries whose `appliesWhen` matches the worktree's enriched-brief, then prompts you to decide which to apply and cite

used by 3 projects (3 invocations)

Use BEFORE creating a pull request — covers title format, body structure, test plan checklist, draft-vs-ready, base/HEAD selection, and how to avoid force-pushing into a PR after review begins.

never used

Use when you are a Claude Code agent spawned by the Mission Control orchestrator for a 12+ hour autonomous run. Establishes session lifecycle, webhook reporting, blocker handling, /compact and memory triggers, mandatory sub-agent dispatch, and pre-completion verification.

used by 8 projects (9 invocations)

Use when editing a Next.js 16 + Vercel app — covers Cache Components configuration, revalidateTag, Drizzle on neon-http (db.batch not db.transaction), zod v4 strict UUID, and Vercel preview deploy verification. Read BEFORE writing code in any Vercel project.

used by 4 projects (4 invocations)

Use when work requires a credential (API token, secret URL, key) and reading process.env returns undefined or the upstream API rejects the value as missing/invalid. Emits a structured marker that Mission Control parses into a credential request for the supervisor.

never used

Use when running tests during a long autonomous task — covers when to run the full suite vs. a single file, how to interpret flaky failures, and the cost/wall-time tradeoffs of full-suite-after-every-edit.

never used

Blocks Bash exposure of 8 known secret envvars and cat .env*. Reads $CLAUDE_PROJECT_DIR/.env files to extract values, scans every Bash command for matches, exits 2 if found.

never fired

Blocks `vercel link --project=<x>` invocations in hierarchical worktrees (those with .vercel/project.json pre-written by the orchestrator). Standalone/greenfield worktrees without the marker are unaffected.

never fired

Verifies a freshly-deployed Vercel preview URL works at runtime before posting agent_done. Returns a strict-format report (PASS/FAIL with reasons). Dispatched via Task tool after vercel deploy.

never dispatched

Evaluate UI quality of a freshly-deployed Vercel preview URL against the brief's design expectations, before the orchestrator posts agent_done. Block-level concerns reject the run; warn-level concerns surface for human review.

never dispatched

Read-only mid-session subagent that searches agent-playbook/canon/ for entries matching a free-text query. Returns the top 3 results with summary + snippet. Caller decides whether to apply and cite.

never dispatched