sit-toolkit 0.20.0

Git-native CLI for versioning AI Skill packages: prompts, schemas, golden tests, and release artifacts.

sit

Git-native versioning for AI Skill packages.

sit adds a semantic layer on top of Git for AI Skills — prompts, schemas, golden tests, runners, and release artifacts. It classifies changes by risk, generates reviewer-ready diffs, and gates commits and releases.

Why sit?

	Pure Git	sit
Prompt change	+13 -2 lines	"Prompt changed (+13 -2); headings: Core Rule, Workflow"
Schema update	raw JSON diff	"breaking" vs "review-required" classification
Golden tests	manual or none	sit test . runs stored cases; --run calls your runner
Version bump	gut feeling	risk-based gate: patch/minor/major suggested by change type
PR review	read every file	sit pr-summary generates structured Markdown

Install

pip install sit-toolkit
sit --version

Python 3.10+ required. Or install from source:

git clone https://github.com/OpenRaiser/SitHub.git
cd SitHub
pip install -e .

Quick Start

# Create a new Skill package
sit init my-skill
cd my-skill

# Or standardize an existing project
cd /path/to/existing-project
sit standardize .

# Validate and test
sit validate .
sit test .

# See what changed
sit diff HEAD~1..HEAD

# Generate a PR summary
sit pr-summary HEAD~1..HEAD

# Release with a version gate
sit release minor . --bundle

Commands

Command	What it does
sit init <name>	Create a new Skill package
sit standardize .	Convert an existing project into a standard Skill package
sit onboard .	Conservatively add sit files to a SKILL.md project
sit doctor .	Check onboarding readiness
sit validate .	Validate manifest, schemas, and golden cases
sit test .	Run golden tests
sit test . --run	Run cases through your configured runner
sit diff A..B	Semantic diff for Git refs
sit diff A..B --prompt	Include prompt/reference text diff
sit pr-summary A..B	Generate PR-ready Markdown
sit report . --compare A..B	Generate Markdown/JSON/HTML report
sit ci-summary . --compare A..B	Generate GitHub Actions summary
sit deps check .	Check deps.yaml dependencies
sit commit -m "..."	Validate/test/version-gate before commit
sit release minor . --bundle	Bump version, tag, and write release bundle

Git passthrough: sit add, sit push, sit pull, sit branch, sit checkout, sit log.

Package Layout

my-skill/
  skill.yaml              # manifest
  prompts/system.md       # prompt files
  schemas/                # input/output JSON schemas
  tests/golden.jsonl      # golden test cases
  scripts/run_case.py     # optional runner
  assets/                 # scanned by semantic diff
  references/             # scanned by semantic diff
  CHANGELOG.md

Semantic Diff Example

Skill Diff
Baseline: paper-webpage-builder@0.3.0
Current: paper-webpage-builder@0.4.0
Risk: review-required
Suggested version bump: minor

[prompt]
  - PROMPT changed skill: SKILL.md -> SKILL.md (+13 -2; headings: Paper Webpage Builder, Core Rule, Workflow)

[script]
  - SCRIPT changed scripts/scan_paper.py (review required; cover with runner or targeted tests)

[reference]
  - REFERENCE changed references/design_principles.md (+27 -3; headings: Design Principles, Fit the Paper)

CI Integration

sit init, sit standardize, and sit onboard create a GitHub Actions workflow:

sit validate "$SIT_PACKAGE_DIR"
sit test "$SIT_PACKAGE_DIR"
sit test "$SIT_PACKAGE_DIR" --run
sit ci-summary "$SIT_PACKAGE_DIR" --compare origin/main..HEAD >> "$GITHUB_STEP_SUMMARY"

Agent Integration

sit can be called by AI agents in three ways:

Python SDK

from sit.sdk import Sit

s = Sit("./my-skill-package")
info = s.info()          # sit.info.v1 contract
test = s.test()          # sit.test.v1 contract
diff = s.diff("./old")   # sit.diff.v1 contract
pr = s.pr_summary("./old")  # sit.pr_summary.v1 contract
report = s.report(compare="./old")  # sit.report.v1 contract

MCP Server

Install with MCP support and run the stdio server:

pip install 'sit-toolkit[mcp]'
sit-mcp-server

Or configure in your MCP client (e.g., Claude Desktop):

{
  "mcpServers": {
    "sit": {
      "command": "sit-mcp-server"
    }
  }
}

Exposes 7 tools: sit_info, sit_validate, sit_test, sit_diff, sit_pr_summary, sit_report, sit_doctor.

LLM Tool-Use Schema

from sit.tool_use import get_tools_openai, get_tools_anthropic

# For OpenAI
tools = get_tools_openai()
response = client.chat_completion(messages=..., tools=tools)

# For Anthropic Claude
tools = get_tools_anthropic()
response = client.messages.create(messages=..., tools=tools)

Research

sit has been validated through multi-agent experiments on AI Skill packaging workflows. See:

• Research proposal (H1-H4)
• Multi-agent mapping
• H2 experiment results and analysis

License

Apache-2.0. See LICENSE.