inizio/groundwork
Structured AI development: orchestrated fan-out, advisor gate, intent routing, and skill-driven workflow for Claude Code
ORCHESTRATOR-ONLY skill. Executor/subagent agents MUST NOT invoke this. Advisor completion gate: mandatory APPROVE/GAPS/CORRECTION/STOP/REPLAN verdict before the orchestrator declares any task complete, plus decision-gate escalation for hard architectural trade-offs.
Mid-session codebase architecture review. All exploration runs in isolated subagents so findings never pollute main context. Surfaces shallow modules, coupling smells, and testability gaps as an HTML report. Use between tasks to review architecture without bloating the session context window.
Resume a motive's in-flight work — reconstruct goal, open slices, and next actions from the motive spine (charter + journal + ledger). If a prior /pause was recorded, agent.last_pause and agent.resume.next_actions are pre-populated — no transcript needed.
Disciplined 6-phase bug diagnosis loop. Build feedback loop, reproduce, hypothesise, instrument, fix with regression test, cleanup with post-mortem. Use for all bugs. Ends with evidence handoff to the orchestrator, who runs the advisor-gate.
Groundwork workflow catalog and entrypoint. Use the specialized Groundwork skills for planning, implementation, diagnosis, testing, pause, and review.
Deletion-first codebase hygiene — deslop (default), deps, lint-debt, docs-staleness, and learnings-staleness modes. Triggers on: deslop, anti-slop, ai slop, cleanup, tidy, housekeep.
Implementation orchestration skill. Decompose into vertical slices and fan out to junior-orchestrators by default; general-purpose is a leaf carve-out only. Validate behavior (not code structure). MANDATORY after a plan or interview. Use vertical-slice skill for conflict-free slice planning and the run ledger (.groundwork/runs/<session_id>.json).
One-question-at-a-time intent-capture interview; produces a motive charter that feeds the planner.
Author and track a motive — a named, persistent work thread that outlives any single session. Captures an Objective, a charter with an open-items register (TBD/TBR), a compiled Decision Log, and named Baselines. Use for any initiative that spans multiple sessions or parallel agents.
Pause an active session — capture the continuation state (pointer, summary, next actions) into the motive spine as a PAUSE event so a later /continue picks up exactly where you left off. The write-side complement to continue.
READ-ONLY cross-artifact coverage map run BEFORE fan-out. Maps acceptance criteria / functional requirements / spec scenarios to slices; flags zero-coverage ACs, untestable ACs, spec↔plan contradictions, negative-scope violations, and missing seams. Use before fan-out on any non-trivial feature, or when advisor plan_soundness is low.
Build throwaway prototypes to flesh out a design before committing. Logic prototypes (interactive TUI) or UI prototypes (variant switcher). Delete when done. Use for spikes and design exploration.
Use when a mistake was corrected 2+ times in a session, a routing or coordination rule was discovered, a streamlined recipe or improvement was found, the user said "remember this", or the struggle-detector surfaced a recurring signal in .groundwork/struggle-signals.jsonl.
Manage structured requirement specifications under doc/specs/ — create, build, verify, and query spec nodes and their traceability data. Triggers on: spec out, write requirements, add requirements, spec upkeep, spec is stale.
Maximum parallel fan-out mode. Slice, write ledger, dispatch all independent slices simultaneously. Triggers on "ultrawork", "ulw", "max fan-out", "go parallel".
Compact Codex bootstrap for the groundwork workflow suite. Load once at session start and again only after context compaction.
Decompose any task into conflict-free parallel slices. Writes a run ledger when the host provides one; otherwise produces an advisory slice table.