New Skill

Create a skill from conversation history or user description.

Steps

1. Detect context

   • If history exists: auto-capture workflow into skill
   • If no history: parse user's description
   • Use thread context clues to infer name, description, and triggers

2. Determine host (Codex, Claude, Cursor) from current runtime

   • Say: "Since I am {host}, I will install it in {host}"

3. Propose the skill name, description, triggers, location, and whether scripts are needed

   • Proceed unless user rejects or corrects

4. Check existing skills for patterns (host-specific)

ls ~/.codex/skills/
ls ~/.claude/skills/
ls ~/.cursor/skills/
ls agents/skills/

1. Create skill structure:

{skill-name}/
  SKILL.md
  resources/         # references, samples, templates
  scripts/           # if tools requested
    run.py|ts|sh

1. Write SKILL.md:

---
name: skill-name
description: one-line
allowed_tools: [list inferred from context]
---

# Skill Name

## When to use
- trigger conditions

## Steps
1. ...
2. ...

## Tools created
- scripts/run.py: what it does

1. Report created files

Flags

--interview: Ask detailed questions (purpose, triggers, inputs, outputs, edge cases)

Script Templates

Python (uv):

def main() -> int:
    return 0

if __name__ == "__main__":
    raise SystemExit(main())

TypeScript (bun):

console.log("ok");

Shell:

#!/usr/bin/env bash
set -euo pipefail
echo "ok"

Rules

• Default to capturing conversation if history exists
• Default host to current runtime and install there
• Ask at most one question, only if ambiguity blocks execution
• Only create scripts if requested
• Match existing skill patterns in repo