new-skill

Create a new custom skill for this project.

What to build: $ARGUMENTS

Steps

1. Gather requirements — If $ARGUMENTS is empty, ask the user:

   • What should the skill do?
   • What slash command name should invoke it (short, kebab-case, e.g. deploy, summarise-pr)?
   • Should only the user be able to invoke it, or can the model also self-activate it?

2. Design the skill:

   • name: short kebab-case identifier matching the slash command
   • description: one precise, verb-led sentence (e.g. "Run …", "Generate …", "Analyse …") — the model uses this for auto-activation, so specificity matters
   • allowed-tools: space-separated, limited to what the skill genuinely needs (Read, Write, Edit, Glob, Grep, Bash)
   • disable-agent-invocation: set true for side-effectful or destructive skills (deploy, commit, send messages); omit for general-purpose skills
   • Body: numbered step-by-step instructions the model follows; reference $ARGUMENTS for the user-supplied target
   • Preprocessors: add !{cmd || echo "fallback"} only when useful context should be captured at activation time (e.g. current git branch, project config); always include a fallback

3. Show the complete draft SKILL.md to the user and ask for approval or edits before writing anything.

4. Write the file to .opencli/skills/<name>/SKILL.md using the write tool — it will prompt for confirmation before touching the filesystem.

5. Confirm success:

   • Report the exact path written
   • Remind the user the skill is discoverable after restarting the session (invoke with /<name> [args])

SKILL.md format reference

---
name: <kebab-case-name>
description: <one sentence — verb-led, specific enough for auto-activation>
allowed-tools: <space-separated: Read Write Edit Glob Grep Bash>
# disable-agent-invocation: true   # uncomment for user-only or dangerous skills
---

<Step-by-step instructions. Use $ARGUMENTS for the user's input.>
<Optional: !{cmd || echo "fallback"} for context injected at activation time.>