Workflow Launchers

Every skill is a launcher: vibecrafted <skill> <agent> dispatches one agent into one workflow and produces a tracked run with a report. Skills share the uniform flag contract (--prompt, --file, --count, --depth, --session), and each also installs a vc-<skill> shortcut.

Canonical pipeline order

scaffold → init → workflow → followup → marbles → audit
        → dou → decorate → hydrate → release

This is the practical build-then-ship order. The full eleven-stage ship cycle (adding implement, review, and polarize) runs as one supervised lifecycle — see Lifecycle overview.

scaffold

Plan architecture from a vague idea. Turns an intention into a measurable, self-sufficient plan that later stages (or a fleet) can execute.

vibecrafted scaffold claude --prompt "I want offline sync for my-app"

init

Orient: history, structural perception, verification. The first context handoff in every repository session — run it before dispatching work.

vibecrafted init claude

workflow

Examine → Research → Implement: a three-phase pipeline that maps the code first, gathers ground truth second, and implements third.

vibecrafted workflow claude --prompt "Plan and implement auth"

followup

Audit post-implementation direction, gaps, and drift. Answers “is this heading the right way and what still feels unfinished” after a delivery.

vibecrafted followup claude --prompt "Check the auth implementation"

marbles

Convergence loop with counterexample elimination: repeated bounded rounds that flood remaining cracks until the gates hold. --count sets the loop count, --depth the plan crawl depth. Control verbs: vibecrafted marbles <pause|stop|resume|session|inspect|delete>.

vibecrafted marbles codex --count 3 --depth 3

audit

Plan-vs-code falsification with a requirements matrix. Read-only: proves or refuses each completion claim against code and test evidence.

vibecrafted audit claude --file plan.md

dou

Definition of Undone audit: a gap analysis across the whole product surface — repo health, install path, docs, discoverability — before release.

vibecrafted dou claude --prompt "Audit launch readiness"

decorate

Visual finishing and UX coherence: detects the existing design language and polishes within it.

vibecrafted decorate codex --prompt "Polish the release surface"

hydrate

Packaging and go-to-market: executes the non-code work that DoU findings surfaced — listings, onboarding, distribution artifacts.

vibecrafted hydrate codex --prompt "Package the product"

release

Ship to production: deployment, publishing, signing, and post-release smoke checks.

vibecrafted release codex --prompt "Prepare release steps"

Additional skills

Skill	Purpose
agents	External fleet delegation contract via agent modes
intents	Plan-to-runtime truth audit
implement	Autonomous end-to-end implementation
partner	Shared steering and executive reasoning with the user
ownership	Full-spectrum operational ownership and delivery
delegate	In-session native subagent delegation
research	Triple-agent research swarm
review	Bounded PR, branch, commit-range, or artifact-pack review
prune	Runtime/publish cone extraction

Examples:

vibecrafted implement codex --prompt "Ship the feature"
vibecrafted intents codex --prompt "Audit what from the plan really landed"
vibecrafted review claude --prompt "Review HEAD~10..HEAD"
vibecrafted ownership codex --prompt "Take the repo from diagnosis to finished surface"
vibecrafted prune codex --prompt "Map what participates in runtime truth"

research is a swarm launcher rather than a single-agent skill: it fires claude + codex + junie on the same questions and collects three independent reports.

vibecrafted research --prompt "Compare storage engines for offline sync"

Use review for a bounded target (a PR, a commit range); use followup when the question is broader direction; use audit when a written plan claims completion and you want it falsified.