Create Local Skill

Create Local Skill

Scaffold a new user-local skill directly into ~/.claude/skills/<name>/ (and optionally ~/.codex/skills/<name>/) so the user can experiment without touching the upstream agentic-skills repo. At the end, offer to promote the skill by copying it into their personal fork of agentic-skills.

Write real directories, not managed roots back to the shared repo. init.sh --uninstall only removes repo-managed skill installs whose source lives inside agentic-skills, so user-authored skills are untouched by upstream sync.

Process

1. Parse $ARGUMENTS for <skill-name> and optional description. If missing, ask the user for:

   • Skill name (kebab-case, e.g. my-skill)
   • One-line description (used in the frontmatter description field — it is how the assistant decides when to invoke the skill, so be specific)
   • Type (execution, analysis, workflow, etc. — default execution)
   • Whether to also create a Codex version in ~/.codex/skills/<name>/ (default yes)

2. Validate the target path:

   • Target is $HOME/.claude/skills/<name>.
   • If the path already exists:
     • If it is a symlink pointing into an agentic-skills repo (readlink contains agentic-skills/base/ or agentic-skills/packs/), refuse — this would shadow a repo-managed skill. Tell the user to choose a different name.
     • If it is a real directory or a user-owned symlink, ask before overwriting.

3. Write the Claude SKILL.md at ~/.claude/skills/<name>/SKILL.md:

   ---
   name: <name>
   description: <description>
   type: <type>
   version: v0.0
   ---
   
   # <Title>
   
   <1-2 sentence purpose statement.>
   
   ## When to use
   
   - <trigger 1>
   - <trigger 2>
   
   ## Process
   
   1. <step>
   2. <step>
   3. <step>
   
   ## Output
   
   <what the skill produces>
   

4. Optionally mirror to Codex at ~/.codex/skills/<name>/SKILL.md with the Codex invocation convention (Invoke as $<name>.). Match the format of existing ~/.codex/skills/*/SKILL.md files.

5. Report the result: show the created paths and remind the user of the runner-specific reload path. Claude Code should run /reload-skills first; /clear starts a new empty-context conversation and can pick up the refreshed registry; restart if the top-level .claude/skills directory did not exist at session start or the skill is still invisible. Codex should start a fresh Codex CLI session if the $ skill list remains stale.

6. Offer to promote. Ask:

   "Would you like to promote this skill to your personal fork of agentic-skills? (y/n)"

   If yes:

   • Ask for the path to their personal fork (e.g. ~/projects/tools/agentic-skills-personal). If they don't have one, explain the options:
     • Fork agentic-skills on GitHub and git clone it.
     • Clone this repo and git remote set-url origin <their-repo-url> to repoint pushes.
   • Verify the path exists and is a git repo (git -C <path> rev-parse --show-toplevel).
   • Copy the scaffolded skill into:
     • <fork>/base/claude/<name>/ (and <fork>/base/codex/<name>/ if a Codex version was created), or
     • <fork>/personal/claude/<name>/ if the user wants to keep personal skills segregated from upstream-syncable dirs. Ask which.
   • If the skill is promoted into a fork's base/ or packs/ tree, also update that fork's tests/harness/bench-coverage.ts and add either a deterministic custom setup under tests/layer4/setups/ or an explicit blocked row with blocked_reason and next_command.
   • For promoted custom setups, include a deterministic quality rubric when practical, or record why quality scoring is blocked/deferred instead of adding a subjective rubric.
   • Recommend pnpm --dir tests bench:coverage before the promoted skill is committed.
   • Show the git status in the fork and suggest a commit message like feat(skill): add <name>. Do not commit or push automatically — leave that to the user.

7. Safety note to surface: pushing to the upstream agentic-skills repo requires write access. A plain clone without write access will be rejected at the remote; a fork pushes to the user's own repo. There is no path by which this skill can push a user's experimental skill to the shared upstream.

Notes

• Never write into this repo's base/ or packs/ directories from this skill. Target paths are always under $HOME/.claude/skills, $HOME/.codex/skills, or an explicitly-supplied personal fork path.
• Do not create documentation files beyond SKILL.md unless the user asks.
• If the user wants version control for their personal skills without a full fork, suggest git init inside a sibling directory and symlinking from ~/.claude/skills/<name> — but default to the simpler "real directory in ~/.claude/skills" flow.

Default Shipping Contract

Follow the shared shipping contract convention in CLAUDE.md.