candango-setup

Files (3)

• SKILL.md
• openai.yaml · agents/
• repo-config-templates.md · references/

SKILL.md (first lines)

# Candango Setup

Scaffold per-repo config consumed by the feature delivery skills.

Borrow Matt Pocock's pattern: keep skills small, compose them through repo-local docs, and make tracker/domain choices explicit before work starts.

## Process

### 1. Explore

Inspect current repo before asking questions:

- `git remote -v`, `.git/config`
- `AGENTS.md`, `CLAUDE.md`
- `CONTEXT.md`, `CONTEXT-MAP.md`, `docs/adr/`
- `docs/agents/`
- `.github/workflows/`, CI config, package scripts, Makefile, task runners
- `.scratch/`, `docs/features/`, existing issue/story docs

### 2. Decide Config

Ask only for values not discoverable from repo.

Required decisions:

- **Issue tracker**: GitHub Issues, Linear, Jira, local markdown, or other.
- **Feature folder layout**: all docs, dashboard HTML, state JSON, design artifacts, and local issues live under `docs/features/<feature-slug>/`.
- **Final validation command**: discovered from CI when possible; otherwise ask: `Is there any command you want me to run at the end of feature development?`
- **Dashboard path**: `docs/features/<feature-slug>/index.html`.
- **Dashboard state path**: `docs/features/<feature-slug>/state.json`.
- **Dashboard cleanup**: `$candango-wrap-up` removes runtime dashboard files before PR when they are not needed.
- **Design artifact path/surface**: default ask per feature; local artifacts go under `docs/features/<feature-slug>/design/`; external options Figma, existing codebase, screenshots, other.
- **Status model**: use `backlog`, `in-progress`, `validation`, `uat`, `done`.
- **Subagent policy**: executor may edit; validator may not edit.
- **Implementation style**: executor subagents use Matt Pocock's `$tdd`.
- **UAT run style**: `$candango-uat-runner` guides one UAT at a time and restarts TDD fix loops on failure.
- **Wrap-up style**: `$candango-wrap-up` creates PRs with caveman descriptions and UAT status.
- **Communication style**: use `$caveman` for all user-facing communication and written reports.
- **Clarification style**: use repo-local `$candango-discover` before planning work when feature terms or business rules are ambiguous.

### 3. Write

Before editing `CLAUDE.md` or `AGENTS.md`, ask explicit permission.

Selection rules:

- If `CLAUDE.md` exists, propose editing `CLAUDE.md`.
- Else if `AGENTS.md` exists, propose editing `AGENTS.md`.
- If neither exists, ask which one to create.

Show:

- target file
- exact block to add/update
- docs files to create

Stop until the user approves. Do not modify `CLAUDE.md`, `AGENTS.md`, or create a new agent instruction file before approval.

Add or update one block:

```markdown
## Autonomous feature skills

### Issue tracker

See `docs/agents/feature-issue-tracker.md`.

### Feature docs

See `docs/agents/feature-docs.md`.

### Design artifacts

See `docs/agents/feature-design.md`.

### Validation

See `docs/agents/feature-validation.md`.

### Autonomous execution

See `docs/agents/autonomous-feature-execution.md`.

### Local support skills

Use `$caveman` for every user-facing message/report and `$candango-discover` for feature clarification against docs.
```

Then create:

- `docs/agents/feature-issue-tracker.md`
- `docs/agents/feature-docs.md`
- `docs/agents/feature-validation.md`
- `docs/agents/feature-design.md`
- `docs/agents/autonomous-feature-execution.md`

Use `references/repo-config-templates.md` for starter content.

### 4. Done

Report config files created and which skills consume them. Tell user these files are editable later; rerun setup only when switching tracker or process.

From lucasbayma/skills, MIT.

More from lucasbayma/skills