skill-runner 0.1.3

Minimal headless agent that executes skill.md files via OpenAI-compatible Chat Completion API

skill-runner

Minimal headless agent (~500 LOC) that executes skill.md files via any OpenAI-compatible Chat Completion API.

Features

• skill.md as system prompt — write instructions in Markdown with optional YAML front matter
• 6 built-in tools — file_read, file_write, file_patch, bash, grep, glob
• Agentic loop — calls API → parses tool_calls → executes → feeds back → repeats
• Zero dependencies — stdlib only (no pip install needed beyond the package itself)
• Any OpenAI-compatible API — OpenAI, Azure, local LLMs, etc.

Install

pip install skill-runner

Quick Start

1. Create a skill file (hello.md):

---
tools: [bash]
max_turns: 3
---
# Hello Skill

You are a friendly assistant. Greet the user and run `echo "Hello!"` to demonstrate tools.

2. Run it:

echo "Hi there" | skill-runner --skill hello.md

Usage

# Pipe input via stdin
echo "Review this code" | skill-runner --skill review.md

# Read input from file
skill-runner --skill review.md --input diff.txt

# Write output to file
skill-runner --skill review.md --input diff.txt --output result.json

# Custom API endpoint and model
skill-runner --skill review.md --input diff.txt \
  --api-base https://api.example.com/v1 \
  --model gpt-4o-mini

# Limit agent loop turns
skill-runner --skill review.md --input diff.txt --max-turns 10

Environment Variables

Variable	Description	Default
SKILL_RUNNER_BASE_URL	API base URL	OpenAI default
SKILL_RUNNER_API_KEY	API key	OPENAI_API_KEY fallback
SKILL_RUNNER_MODEL	Model name	gpt-4o

CLI flags (--api-base, --model) override environment variables.

skill.md Format

A skill file is Markdown with an optional YAML front matter block:

---
tools: [file_read, bash, grep]
max_turns: 20
---
# Your Skill Name

System prompt instructions go here...

Front matter fields:

Field	Type	Default	Description
tools	list	all 6 tools	Which tools to enable
max_turns	int	50	Max agent loop iterations

Available tools: file_read, file_write, file_patch, bash, grep, glob

GitHub Actions Example

Use skill-runner for automated code review in CI:

- name: Install skill-runner
  run: pip install skill-runner

- name: Run AI review
  run: |
    git diff origin/main...HEAD | skill-runner \
      --skill skills/code-review.md \
      --output /tmp/review.txt
  env:
    SKILL_RUNNER_API_KEY: ${{ secrets.LLM_API_KEY }}
    SKILL_RUNNER_MODEL: gpt-4o-mini

See examples/ai-review.yml for a complete workflow.

License

MIT