aidlc-init
Start an AI-DLC workflow — run the whole Initialization phase (mint the intent, detect the workspace, build state) in one step, without typing a stage. The engine normally auto-creates the first inten
SKILL.md.claude/skills/aidlc-init/SKILL.md
GitHubAI-DLC - start a workflow (create the first intent)
Start a fresh AI-DLC workflow. The workspace shell ships in dist/ (no setup
command), and the engine auto-creates the first intent when you describe what to
build - this skill is opt-in packaging over that creation move. Initialization is a
PHASE, not a single stage — it mints the intent, detects the workspace
(greenfield/brownfield), and builds aidlc-state.md together, in one
deterministic call. There is no per-init-stage runner because an init stage has
no standalone meaning.
Steps
Read the user's
$ARGUMENTS. The recognized flags are--scope <name>,--depth <level>, and--test-strategy <level>; the rest is a freeform description of what to build.The user named a scope with
--scope <name>: create the intent (run the initialization phase). Forward the recognized flags as-is, and pass any freeform description text via--arguments "<text>"(intent-createreads the description from the--argumentsflag, NOT a positional: forwarding it bare would silently drop it). ALSO derive a short--label: a 2-3 word kebab-case essence of what's being built ("I would like to build a simple calculator application"gives--label "simple calc"). The label becomes the readable, date-prefixed record dir name (<YYMMDD>-simple-calc); the full--argumentstext is preserved separately in the audit + state. When only a scope was supplied, omit--argumentsand--label(the tool then falls back to the scope token):aidlc engine intent create --scope <scope> --arguments "<description>" --label "<2-3 word essence>"Print the tool's output and stop. This does not advance a stage; run
/aidlcafterwards to continue.The user described the work but gave no
--scope(even when the description starts with a scope name, as infeature flags for billing): do not create it on a scope they have not seen. Pass the arguments to the engine as new work; it proposes the plan that fits the description (for examplebugfixfor a described bug) or offers to compose one, and asks the user to choose. Work already in progress is left as it is:aidlc engine orchestrate next --new-intent $ARGUMENTSBefore acting on each directive, read
.claude/aidlc-common/protocols/stage-protocol.mdonce per session, then every.claude/aidlc-common/protocols/stage-protocol-<module>.mdnamed bydirective.protocol_modules. Act on the directive exactly as theaidlcskill's forwarding loop describes. From here the flow IS the/aidlcflow - continue its loop until the directive says stop.If the user gave neither a scope nor a description, do not run a bare
intent-create: ask what they want to build or which scope to use.