SkillCast

Write Skills once, deliver everywhere. — A universal Skill format converter for AI agent platforms.

Quick Demo

$ pip install skillcast

$ skillcast list
📥 Input formats:
  - generic   - claude   - hermes   - cursor
📤 Output formats:
  - claude    - hermes   - cursor   - codex

$ skillcast init java-interview
✅ Created: java-interview.yaml

$ skillcast convert java-interview.yaml --all -o output
✅ Parsed: java-interview — Helps prepare for Java technical interviews
  📄 output/java-interview.claude.md
  📄 output/java-interview.hermes.md
  📄 output/java-interview.cursor.cursorrules
  📄 output/java-interview.codex.json
🎉 Done! Output in: output/

$ ls -1 output/
java-interview.claude.md
java-interview.codex.json
java-interview.cursor.cursorrules
java-interview.hermes.md

💡 Run bash demo/demo.sh to see the full workflow live in your terminal.

Installation

pip install skillcast
# or zero-install:
uvx skillcast --help

Supported Platforms

Platform	Input	Output
Claude Code	SKILL.md (YAML frontmatter)	SKILL.md
Hermes	SKILL.md (with hermes metadata)	SKILL.md
Cursor	.cursorrules (plain text)	.cursorrules
Codex CLI	— (via generic YAML)	JSON config
Generic	YAML / JSON	—

CLI Reference

skillcast list                          List supported input/output formats
skillcast init <name>                   Create a new Skill template
skillcast convert <file> --all          Convert to all target platforms
skillcast convert <file> --to <fmt>     Convert to a single platform

Library Usage

from skillcast import parse_skill, generate, generate_all

# Parse any Skill file
ir = parse_skill("my-skill.yaml")

# Generate for a specific platform
claude_output = generate(ir, "claude")

# Generate for all platforms
outputs = generate_all(ir)

Architecture

Parser → IR (SkillIR dataclass) → Generator

Each platform has an isolated Parser and Generator, operating on a shared normalized intermediate representation.

License

MIT — see LICENSE