agent-atelier

Candidate pipeline lifecycle — enqueue work items as a candidate set for validation, activate the next queued set into the exclusive validation slot, or clear the active set after completion or demotion. Supports single and batch candidates with fate-sharing semantics. Use when a builder finishes implementation, when the orchestrator needs to start validation, or when validation passes or fails. Triggers on 'candidate', 'enqueue candidate', 'activate candidate', 'clear candidate', 'promote to candidate', 'candidate queue', 'next candidate', 'demote candidate', 'candidate ready', 'submit for validation', or 'validation slot'.

Work item execution lifecycle — claim a work item to start implementation, send heartbeats to keep the lease alive, requeue if stuck, mark complete with evidence, or record an attempt. Use when an executor needs to start work, renew a lease, return work to the queue, record a failed attempt, or finalize a work item. Triggers on 'claim', 'heartbeat', 'requeue', 'complete', 'record attempt', 'I'm done with WI-NNN', 'start working on', or 'mark as done'.

Human decision gate lifecycle — list open gates, create a new gate for user decisions, or resolve an existing gate. Use when the orchestrator needs to escalate a decision to the user, when viewing pending decisions, or when the user responds to a gate. Triggers on 'gate', 'human gate', 'escalate', 'decision needed', 'open gate', 'resolve gate', 'HDR', 'pending decisions', or when the 3-test criteria (irreversibility, blast radius, product meaning) indicate a human decision is required.

Initialize the orchestration workspace with default state files and directory structure. Use when bootstrapping a new project, when orchestration state is missing or corrupt, when the user says 'init', 'initialize', 'bootstrap', 'set up the workspace', 'start a project', 'prepare for agent-atelier', or 'create state files'. Also triggers automatically before /agent-atelier:run if .agent-atelier/ does not exist. Safe to re-run — only creates files that don't already exist.

Background monitor lifecycle — spawn continuous monitors, poll for events, stop monitors, or check health. Use when starting the orchestration loop, polling for state changes, cleaning up on exit, diagnosing monitor health, or tracking CI runs. Triggers on 'spawn monitors', 'check monitors', 'stop monitors', 'monitor status', 'spawn ci monitor', 'poll events', 'respawn', 'dead monitor', 'background processes', 'ci monitor', 'start monitors', 'kill monitors'.

Start the autonomous development loop -- spawn the agent team, begin the orchestration state machine, and drive work items through the full lifecycle from DISCOVER to DONE. Use when the user says 'run', 'start', 'begin', 'launch the loop', 'start the team', 'run the development loop', 'go', 'pick up where we left off', or 'kick it off'. This is the entry point for the entire agent-atelier system.

Show the orchestration dashboard — current loop mode, work item summary, active candidate set, candidate queue, open gates, and next action. Use when checking progress, understanding what's happening, reviewing the backlog, recovering from a crash, or at the start of any session. Triggers on 'status', 'show state', 'what's going on', 'dashboard', 'where are we', 'progress', 'overview', 'how's it going', 'what's next', or 'orchestration report'. The go-to skill for situational awareness.

Validation evidence recording — register a validation run manifest with check results and link it to the work item. Use when a validator finishes a validation run, when recording test evidence, or when the orchestrator needs to formally register a validation outcome. Triggers on 'validate', 'record validation', 'register run', 'validation complete', 'validation failed', 'record evidence', 'VRM done', or 'validation manifest'.

Health check and mechanical recovery — detect stale leases, expired candidates, stuck reviews, and blocked work items, then take safe recovery actions. Also enforces operating budgets and replays interrupted transactions. Use for routine health checks, when something seems stuck, or when the user says 'watchdog', 'health check', 'check for stale items', 'anything stuck?', 'run maintenance', 'check budgets', 'recovery sweep', or 'clean up orphaned leases'. Also appropriate after a session crash or long idle period.

Work item planning — list all work items, show details of one, or create/update work items. Use for backlog management, viewing work item details, creating new work items, or updating existing ones. Triggers on 'work item', 'WI', 'backlog', 'list items', 'show WI-NNN', 'create work item', 'update work item', or 'upsert'. For execution-phase operations (claim, heartbeat, complete), use the execute skill instead.

7 hooks across 7 events