cmdop-skill 2026.4.7.2

Build and publish CMDOP skills

cmdop-skill

Build and publish CMDOP skills.

Install

pip install cmdop-skill

Quick Start

Scaffold a new skill

cmdop-skill init

Interactive wizard: name (with PyPI availability check), description, author.

Or write from scratch

from cmdop_skill import Skill, Arg

skill = Skill()  # name, version, description from pyproject.toml

@skill.command
async def greet(name: str = Arg(help="Who to greet", required=True)) -> dict:
    """Say hello."""
    return {"message": f"Hello, {name}!"}

if __name__ == "__main__":
    skill.run()

Skill() auto-resolves name, version, description from the nearest pyproject.toml. Explicit arguments still win.

Core Features

Feature	Description
@skill.command	Turn async/sync functions into CLI subcommands
Arg()	Declarative arguments: help, required, default, choices, nargs, action
@skill.setup / @skill.teardown	Lifecycle hooks (run before/after every command)
SkillCache	Typed disk cache with TTL, scoped per skill — docs
TestClient	Test harness: client.run("cmd", key=val) / client.run_cli("cmd --flag")
JSON output	All commands return {"ok": true, ...} — exit 0 on success, exit 1 on error
Auto src/ path	src/ added to sys.path automatically
Single source of truth	name, version, description from pyproject.toml, no duplication

SkillCache

Built-in typed disk cache — no extra dependencies, platform-aware paths:

from cmdop_skill import SkillCache

cache = SkillCache("my-skill")
cache.set("data", payload, ttl=86400)   # 24 hours
data = cache.get("data")                # None if missing or expired

Paths: ~/Library/Caches/cmdop/skills/<skill>/ (macOS), ~/.cache/cmdop/skills/<skill>/ (Linux). Full reference → @docs/cache.md

Skill Structure

my-skill/
├── pyproject.toml
├── Makefile
├── README.md
├── skill/
│   ├── config.py       # SkillConfig manifest (required)
│   └── readme.md       # description shown in marketplace
├── src/my_skill/
│   ├── __init__.py
│   └── ...
└── tests/

skill/config.py — minimal manifest:

from cmdop_skill import SkillConfig

config = SkillConfig()  # all fields auto-resolved from pyproject.toml

CLI

Command	Description
init [path]	Scaffold a new skill project
install <path>	Symlink skill into system skills directory
uninstall <name>	Remove skill
bump [--minor|--major]	Bump version in pyproject.toml
check-name <name>	Check PyPI availability
release [--no-bump|--no-publish]	Bump + build + PyPI + CMDOP
publish [--path]	Publish to CMDOP marketplace
list	List your published skills
config set-key / show / reset	Manage API key

API key resolution order: --api-key flag → CMDOP_API_KEY env → saved config → interactive prompt.

Python API

Export	Description
Skill	Main class — registers commands, runs CLI
Arg	Argument descriptor
SkillConfig	Typed skill manifest (Pydantic model)
SkillCache	Disk cache — see @docs/cache.md
CacheEntry	Pydantic model for a cached value with metadata
TestClient	Test harness
resolve_project_meta	Resolve name/version/description from pyproject.toml
generate_manifest	Generate skill/config.py content
publish_skill	Programmatic publish
json_output / wrap_result / format_error	Output helpers
scaffold.ScaffoldConfig / scaffold.scaffold_skill	Programmatic scaffolding

Developer Docs

Deep-dive references in @docs/:

• Architecture — system diagram, components, data flow
• Publish Flow — CLI → Django → DB
• Install Flow — deep link → Go loader → validation
• Package Environment — venv, PATH enrichment
• Skill Format — skill.md frontmatter rules
• SkillCache — cache API, paths, testing