Spec-Driven Development orchestrator — scaffold .sdd/, create/select a feature, show pipeline status, and route to the next phase. Triggers: sdd, spec-driven, start sdd, sdd status, sdd init, new feature spec, what phase, which sdd phase, spec kit, speckit.
Installation
Installer avec Codex ou Claude Copiez ce prompt, collez-le dans Codex, Claude ou un autre assistant, puis laissez-le vérifier la page du skill et l'installer pour vous.
Spec-Driven Development orchestrator — scaffold .sdd/, create/select a feature, show pipeline status, and route to the next phase. Triggers: sdd, spec-driven, start sdd, sdd status, sdd init, new feature spec, what phase, which sdd phase, spec kit, speckit.
SDD — Orchestrator
TL;DR — Scaffold .sdd/ (1) → create or select a feature (2) → show status & route to the next m-sdd-<phase> skill (3). Read _shared.md for the rules every phase shares.
This is the entry point. Phase skills (m-sdd-constitution … m-sdd-tasks-to-issues) do the per-phase work; this skill sets up state and tells you where you are.
Phase 0 — Preflight
Read sibling m-sdd/_shared.md.
Require jq: command -v jq >/dev/null || echo "MISSING jq" — if it prints
MISSING jq, STOP: tell the user to install it (brew install jq) and do
not run any sdd_* function until it is present.
source <skills-root>/m-sdd/sdd-lib.sh; root="$(pwd)" (<skills-root> is defined in _shared.md)
Phase 1 — Scaffold (idempotent)
sdd_scaffold "$root" — creates .sdd/{config.json,state.json,templates/} and specs/.
Mention that .sdd/ and specs/ should be committed to the project repo.
Phase 2 — Feature selection
First show what exists: sdd_list "$root" (one line per feature, * marks active). Then interpret intent:
New feature ("new "): id="$(sdd_create_feature "$root" "<name>")" then report $id (this also makes it active).
Select existing: sdd_set_active "$root" "<id>" — never hand-edit config.json.
Just status: skip to Phase 3.
Phase 3 — Status & route
sdd_status "$root" — prints the active feature, each phase's status, and Next:.
Tell the user which skill to run next: m-sdd-<next> (from sdd_get_next).
If Next: constitution and no .sdd/constitution.md, recommend m-sdd-constitution.
After the constitution, mention the optional m-sdd-knowledge (builds .sdd/knowledge.md — project facts that ground specify/clarify); it is not a tracked phase, so it won't appear in sdd_status.
After specify, mention the optional m-sdd-checklist (requirement-quality checklists in specs/<id>/checklists/ — "unit tests for English"); also untracked.
If the user has UI designs made elsewhere (e.g. in an AI chat tool), note they can drop the HTML into specs/<id>/design/ — m-sdd-specify/m-sdd-plan will index and reference it (visual reference, not a tracked phase).
Notes
Never hand-edit state.json; always go through sdd-lib.sh.
Pipeline order lives in one place: sdd_phases in sdd-lib.sh (surfaced via sdd_status/sdd_get_next) — don't duplicate it.
phase-instructions.md files in the skill folders are install-time input for non-Devin adapters (consumed by install.sh), not runtime reading.