| name | add-skill |
| description | Scaffold a new skill in this ai-config repo: create the SKILL.md, update the README table and mermaid diagram, then run setup. Use when: adding a new skill, creating a skill, scaffolding a SKILL.md in this repo. |
| argument-hint | Skill name (kebab-case) and a short description of what it does |
Add Skill
Scaffold a new global skill in this repo and wire it into all the places that must stay in sync. This skill is itself project-local — its source lives in .agents/skills/ (AI-agnostic) and is symlinked into .claude/skills/; it is not linked into the global skill set.
Step 1: Gather the skill details
You need:
- Name — kebab-case, matches the directory (e.g.
read-pr)
- Description — the
description: frontmatter, including the Use when: trigger phrases that help the agent know when to invoke it
- Argument hint (optional) — what the skill takes as input
- Building block? — does it compose other skills, get composed by them, or stand alone?
If any of these are unclear from the request, ask before scaffolding.
Step 2: Create the SKILL.md
Create skills/<name>/SKILL.md (note: the repo's skills/ dir, not .claude/skills/, so setup links it globally).
Follow the structure of existing skills:
- YAML frontmatter with
name, description, and optional argument-hint
- A title heading and a one-line purpose
- Numbered
## Step N: sections for the procedure
- Match the voice and concision of the existing skills (imperative, no filler)
If the skill needs helper scripts, put them in skills/<name>/scripts/ and add a chmod +x line for them to setup under the "make sure scripts are executable" section.
Step 3: Update the README table
In README.md, add a row to the skills table (## AI Skills). Keep the row order sensible — group it near related skills. Mark it as a building block in the description if it's meant to be composed by other skills.
Step 4: Update the mermaid diagram
In the same README, update the How the skills relate mermaid diagram:
- If the skill composes others or is composed, add it to the appropriate
subgraph (workflows or blocks) and draw the --> edges.
- If it's standalone, add it to the
standalone subgraph and the prose line below the diagram.
Step 5: Run setup
Run the setup script so the new skill is installed into ~/.agents/skills/ (and the ~/.claude/skills/ symlink). This skill runs inside the ai-config repo, so resolve the script from the repo root rather than hardcoding a path:
"$(git rev-parse --show-toplevel)/setup"
Confirm the output shows <name>: linked.
Step 6: Summarize
Report what was created and changed:
- The new
skills/<name>/SKILL.md path
- The README table row and diagram edges added
- Confirmation that setup linked the skill
Remind the user the skill won't appear in an already-running Claude Code session until it's restarted.