Agent Skill Infrastructure

From dev to ops: quality check, behavior test runner, and version awareness for Agent Skills.

Install

# PyPI (recommended)
pip install agent-skill-infra

# With LLM judge support (requires Anthropic API key)
pip install agent-skill-infra[llm]

# From source
git clone https://github.com/Liber1917/agent-skill-infra.git
cd agent-skill-infra
uv sync

CLI Commands

skill-test — Behavior Test Runner

Run evals.json test suites against skills with multi-dimension judgment.

# Run tests
skill-test run tests/fixtures/evals.json

# Output as JSON
skill-test run tests/fixtures/evals.json --output json

# Use mock adapter (no Agent runtime needed)
skill-test run tests/fixtures/evals.json --adapter mock

Judge types: keyword, schema, llm, flow, snapshot

skill-quality — Quality Assessment

Score your SKILL.md against helloandy 8-dimension framework.

# Quick quality check
skill-quality /path/to/skill/SKILL.md

# JSON output
skill-quality /path/to/skill/SKILL.md --output json

# With agent-skill-linter integration
skill-quality /path/to/skill/SKILL.md --lint

# With security scan (requires cisco-scanner)
skill-quality /path/to/skill/SKILL.md --security

skill-version — Version Awareness

Track changes, detect regressions, and roll back safely.

# See what changed between versions
skill-version diff /path/to/skill --old-ref HEAD~3 --new-ref HEAD

# Diff + security analysis
skill-version check /path/to/skill --security

# Roll back to previous version
skill-version rollback /path/to/skill --target-ref HEAD~1 --yes

# Store and compare baselines
skill-version baseline store /path/to/skill case-1 output.txt
skill-version baseline detect /path/to/skill case-1 output.txt

Modules

Development

# Install dev dependencies
uv sync

# Run tests (162+)
uv run pytest

# Lint & format
uv run ruff check .
uv run ruff format .

# Type check
uv run pyright

# Build
uv build

License

MIT