ai-agents-workflow

Add an entry (skill, plugin, baseline, etc.) to <artifact-root>/config/PROJECT_CONFIG.md.

Resume an in-progress task by task_id, or show a menu of resumable tasks.

Bootstrap <artifact-root>/config/PROJECT_CONFIG.md for this repo via the init agent.

Harvest review comments from a GitHub PR into <artifact-root>/knowledge/pr-lessons.md so the same mistake isn't made twice.

Remove an entry from <artifact-root>/config/PROJECT_CONFIG.md.

MUST BE USED for every new multi-step user task. Coordinates intake classification, routing to specialists (lead/executor/reviewer/design-agent/integration-checker/delivery-pm), enforces P1/P2/P4 user-approval gates, runs the review loop, and finalizes workflow artifacts. Invoked automatically by `/ai-agents-workflow:task`.

Delivery planning specialist for decomposition, blockers, reprioritization, and scope split proposals.

Conditional mobile UX specialist for flows, usability, CTA hierarchy, and loading/error/empty states.

Generic executor. Implements approved subtasks in any domain and emits an Implementation Report. Stack-agnostic; domain skills/plugins/baselines resolve from <artifact-root>/config/PROJECT_CONFIG.md at runtime.

Generate or maintain <artifact-root>/config/PROJECT_CONFIG.md for the consumer repo via scoped discovery, multiple-choice questions, and a review-and-comment gate. Modes - init, update, add, remove.

Owns dispatch orchestration — subtask skeleton, pre-dispatch checklist, post-dispatch artifact gate, clarifying-questions hold, rework routing, and token-saving rules. Use before EVERY agent dispatch (except direct-answer and Delivery PM task-level intake) and after every return. Bundle composition itself is delegated to context-minimizer.

Classify an incoming task into one of five paths (direct-answer, plan-only, execution-trivial, execution-simple, execution-full) using checklist-based rules, then confirm the chosen path with the user via a mandatory radio-button popup before any pipeline work begins. Use at Step 0 before any artifact creation.

Owns the orchestrator state files (`orchestration-state.json` hot + `orchestration-history.json` history) — schemas, field semantics, phase/stage transitions, history-consistency contract, and post-approval closure. Use before every subtask transition, on first state write, and at post-approval closure. Read by orchestrator-dispatch (pre-dispatch hot-state read) and orchestrator-user-gates (P4 history validation).

Format spec for per-agent telemetry lines and Context Manifest subsections in `<subtask_id>/summary.md`. Use when specifying agent output requirements and when validating dispatch returns against the expected schema. Task-level aggregation (writing the rolled-up `<task_id>/summary.md`) is delegated to telemetry-summary.

Pauses for user input at four orchestration gates via AskUserQuestion. **P1** — fires after Delivery PM appends `<!-- section:delivery-plan -->` and before any subtask agent dispatch (hard-blocked by `pre-task-guard.js`). **P2** — fires when phase-N subtasks are all closed and phase-N+1 has pending work. **P4** — fires when all subtasks closed, gates empty, before setting `phase: complete`. **P5** — fires after completion for retrospective (skipped for ≤2-subtask all-first-cycle tasks). Use when reaching any of these checkpoints.

11 hooks across 4 events