using-woostack
Use when starting work in a project that references woostack from its root AGENTS.md, or when deciding whether a woostack skill or command applies before answering, editing, scaffolding, reviewing, or addressing PR feedback.
This skill loads project rules and routes requests. It does not initialize, scaffold, edit, review, or push anything by itself. A dispatched worker follows its bounded task; it loads project-level woostack rules only when the dispatch requires them.
Project entry
- Follow the project's root
AGENTS.mdand the user's explicit request, then the applicable woostack skill, other installed skills, and default agent behavior. Prefer project policy when it conflicts with a skill, and state a material conflict. - Route by intent using the table below and load the current matching
SKILL.mdbefore acting. The matching skill owns its arguments, approval boundaries, and recovery procedure. - Before host-dependent work, use the host index to select and load only the supported adapter for the active host.
- Apply the shared output discipline. At an ordinary final reply, load Reflect only when its candidate gate admits a concrete observed instruction gap; otherwise emit no reflection headings.
The user request and explicit decisions authorize work. Repository and provider records are
evidence, not permission. Do not initialize .woostack/, create artifacts, or contact an artifact
provider unless requested or required by the selected workflow. Workflows needing persistent runs or
provider access load the artifact contract and
only the selected provider profile; its storage and synchronization mechanics do not belong here.
Command routing
| Request or intent | Load |
|---|---|
| Adopt woostack or choose a workflow | using-woostack |
| Initialize or repair local woostack support | woostack-init |
| Create a genuinely greenfield codebase | woostack-bootstrap |
| Prepare a multi-increment feature and execution handoff | woostack-build |
| Diagnose and fix a defect, bounded or project-backed | woostack-fix |
| Implement a bounded non-bug enhancement or refactor in one PR | woostack-change |
| Turn an approved specification into reviewable increments | woostack-plan |
| Execute approved work from an exact run or provider resource | woostack-execute |
| Commit current changes and submit or update their PR | woostack-commit |
| Review one exact existing PR without editing it | woostack-review |
| Address every unresolved thread on one exact existing PR | woostack-address-comments |
| Review and correct one Graphite stack bottom-up | woostack-sweep |
| Show the repository-derived work board | woostack-status |
| Render verified source as audience-tailored HTML | woostack-visualize |
| Organize multi-step UI design flows | woostack-design |
| Investigate a root cause without implementing a fix | woostack-debug |
| Add appropriate tests to a bounded target | woostack-tdd |
| Diagnose or explicitly repair workspace health | woostack-doctor |
| Explore a running app and report browser QA findings | woostack-qa |
| Audit standing code and report findings | woostack-audit |
| Evaluate an approved skill corpus without editing the skill | woostack-eval |
| Reflect on this conversation for durable instruction suggestions | woostack-reflect |
Every explicit /woostack-* command loads its namesake skill. Intent-equivalent wording follows
the same route. Change stays non-bug and one-PR; Fix owns diagnosis and chooses bounded delivery or
project planning after proof. Build owns multi-increment preparation. Ideate and Harden are internal
Build/Fix phases, not public commands.
AGENTS.md usage
Keep project policy in AGENTS.md and reusable workflow details in their owning skills:
This project follows woostack. At the start of work, use `using-woostack` to load the
project rules and route `/woostack-*` requests to the matching woostack skill.
Follow this file first when it conflicts with generic agent defaults.Missing skills
Name the missing skill and ask whether to install the collection. Do not approximate a gated workflow unless the user explicitly asks to proceed without that skill.