extend

/aura-frog:extend

Create project-level skills / rules / commands. Always writes to .claude/ in the user's project — NEVER to the plugin's aura-frog/ folder.

---

Usage

/aura-frog:extend propose <kind> <description>           # interview-driven skeleton; no write yet
/aura-frog:extend create  <kind> <name>                  # write file from approved skeleton
/aura-frog:extend list                                   # show project extensions
/aura-frog:extend remove <kind> <name>                   # delete a project extension (confirm required)

<kind> ∈ skill | rule | command.

Protocol — propose <kind> <description>

1. Verify project context. Refuse if no .claude/CLAUDE.md exists ("not an Aura Frog project; run /project init first").
2. Read project context — .claude/CLAUDE.md, recent prompts/turns, file tree. Ground the proposal in actual usage.
3. Draft a skeleton with the right frontmatter for the kind:
   • skill → name, description, when_to_use, autoInvoke (default false), user-invocable: false, allowed-tools, effort
   • rule → priority (Critical/High/Medium), tier (core/agent/workflow), applies_to
   • command → usage block, protocol (imperative), tie-ins
4. Show the skeleton to the user with the target path printed.
5. Wait for explicit confirmation. y / yes / looks good / create it advances to step 6. Anything else aborts.
6. Run /aura-frog:extend create <kind> <name> with the approved skeleton.

Protocol — create <kind> <name>

1. Resolve target path based on kind:
   • skill → .claude/skills/<name>/SKILL.md
   • rule → .claude/rules/<tier>/<name>.md (tier from frontmatter)
   • command → .claude/commands/<name>.md
2. Refuse if path collides with aura-frog/ — explicit error: "/aura-frog:extend writes only to .claude/, never to plugin folder".
3. Refuse if file already exists — suggest propose to draft an update instead.
4. Create parent directory if missing (.claude/skills/<name>/ etc.).
5. Write the approved skeleton verbatim.
6. Run reference integrity audit (per CLAUDE.md "LESSON LEARNED — Reference Integrity Rule"):
   • For new rules: confirm ≥1 inbound reference will exist (or output: "rule created but currently orphaned — wire it from a skill/agent")
   • For new skills: confirm description triggers don't shadow an existing plugin skill
   • For new commands: confirm name doesn't collide with a plugin command
7. Append .claude/extensions.log (project-local, gitignored by default):

   {"ts":"<ISO>","kind":"skill","name":"<name>","path":"<rel-path>","reason":"<one-line>"}
   

8. Render the post-create summary: kind, path, audit findings.