toolkit-skill-author

# toolkit-skill-author ## Pre-load ```bash !test -f CONTRIBUTING.md || (echo "CONTRIBUTING.md missing — run from repo root." && exit 1) ``` **Verify `skill-creator` is callable in this session.** Scan the available-skills list for a skill named `skill-creator` (bare or namespaced, e.g. `skill-creator:skill-creator`). If absent, abort with **exactly this message**: > `skill-creator` plugin is required. Run `/plugin install skill-creator@claude-plugins-official`, enable it, then restart this session and re-run `/toolkit-skill-author`. Do **not** fall back to manually scaffolding SKILL.md — the handoff is core to the workflow. Load authoritative context: ```bash !cat CONTRIBUTING.md !bash .claude/skills/toolkit-skill-author/scripts/find-peers.sh skills ``` - @.claude/skills/toolkit-skill-author/references/decision-rules.md - @.claude/skills/toolkit-skill-author/references/handoff-preamble.md - @.claude/skills/toolkit-skill-author/references/registration-checklist.md ## Phase 0 — Parse intent and apply gates From the contributor's initial prompt, extract (without asking yet): - **`target_name`** — did they name a specific skill (e.g., *"extend `monitoring-advisor`"*, *"create a skill called `foo-bar`"*)? - **`action`** — `extend`, `new`, or `unknown`. - **`bucket`** — did they state a capability bucket? Apply the gates below **before** any survey question. Each gate halts when it fires. ### Gate A — Forbidden bucket If `bucket` is `Agent-routing` (or a synonym like ...