v0.8.0 — generate + watch + doctor

Three wow features

aif generate — skill from plain English

pip install anthropic  # optional dependency
aif generate "code review skill for React PRs" -o code-review.md

Calls the Anthropic API, produces a deploy-ready SKILL.md that passes all quality checks. Post-processes with aif fix, validates with aif check, retries on failure. One command → A+ skill.

aif watch — live terminal grade

aif watch my-skill.md

Persistent terminal display that updates instantly on every file save. Color-coded grade (green/yellow/red), progress bar, lint/security counts, running timer. Makes skill authoring feel like a game.

aif doctor — repo health report

aif doctor .

Scans your entire repo for skills, grades each A+..F, prints a sorted report with actionable fix suggestions. One command → full picture.

Full command list (v0.8.0)

Command	What
aif check FILE	lint + security scan
aif score FILE	letter grade + badge
aif fix FILE	autofix lint issues
aif scan FILE	security only
aif lint FILE	lint only
aif generate "..."	LLM-generate a skill
aif watch FILE	live grade display
aif doctor DIR	repo-wide report
aif mcp-server	MCP for Claude Desktop

Install

pip install aif-skillforge   # cross-platform wheels

106 tests passing across all Python modules.

v0.7.0 — Autofix + MCP server + Leaderboard

Three new features

1. aif fix — autofix SKILL.md

aif fix my-skill.md --write     # fix and overwrite
aif fix my-skill.md --check     # dry-run, exit 1 if changes needed (CI)
aif fix my-skill.md --diff      # show unified diff

Mechanically fixes 5 lint rules: NameFormat (kebab-case), missing frontmatter, oversized descriptions, missing required sections, empty blocks. Takes a broken skill from parse-error to A+ in one command.

2. aif mcp-server — MCP for Claude Desktop / Cursor

{
  "mcpServers": {
    "skillforge": {
      "command": "aif",
      "args": ["mcp-server"]
    }
  }
}

4 tools: check_skill, score_skill, scan_skill, fix_skill. Any MCP client can invoke SkillForge natively.

3. Weekly skills leaderboard

leaderboard/LEADERBOARD.md — GitHub Action scans for public SKILL.md files, grades with aif score, publishes ranked top-25 weekly.

Install

pip install aif-skillforge   # cross-platform wheels
aif fix SKILL.md --write
aif score SKILL.md

Wheels

Linux x86_64/aarch64, macOS x86_64/arm64, Windows x64, source distribution.

v0.6.4 — Correct cross-platform wheels

Fixes the Linux wheel tag + Windows build

v0.6.2 and v0.6.3 had wheel-tagging issues that prevented installation on Linux x86_64 and Windows. v0.6.4 ships correctly-tagged wheels for all 5 platforms:

• cp39-abi3 manylinux_2_17 x86_64 (fixed in this release)
• cp39-abi3 manylinux_2_17 aarch64
• cp39-abi3 macOS 10.12+ x86_64
• cp39-abi3 macOS 11.0+ arm64
• cp39-abi3 Windows x64

All wheels are stable-ABI (abi3-py39): one wheel per platform works on Python 3.9+, forward-compatible with 3.14+.

Install

pip install aif-skillforge

GitHub Action

- uses: LiqunChen0606/skillforge@v0.6.4

Pre-commit

- repo: https://github.com/LiqunChen0606/skillforge
  rev: v0.6.4
  hooks:
    - id: aif-check

What's in v0.6.x

• aif check — 10 lint + 6 OWASP AST10 security checks
• aif score — letter grade A+..F + Shields.io badge + SVG
• aif scan — security-only output
• aif lint — lint-only output
• Pre-commit hooks: aif-check, aif-lint, aif-scan, aif-score
• Claude Code plugin with /lint-skill, /scan-skill, /score-skill, /sign-skill, /verify-skill

v0.6.2 — Cross-platform wheels

Prebuilt wheels for Linux, macOS, Windows

v0.6.2 ships wheels for every common CI platform, so pip install aif-skillforge just works without a Rust toolchain:

• Linux x86_64 (manylinux2014)
• Linux aarch64 (manylinux2014)
• macOS x86_64 (10.12+)
• macOS arm64 (11.0+)
• Windows x64
• Source distribution (builds on any platform with Rust)

What changed

• Switched to PyO3 abi3-py39: one wheel works on Python 3.9+, forward-compatible with future versions
• New publish-wheels.yml GitHub Actions workflow builds + uploads wheels on tag push
• Dropped Rust fallback from the SkillForge GitHub Action (no longer needed)

Same feature set as v0.6.0 / v0.6.1

• aif check SKILL.md — 10 lint + 6 security checks
• aif score SKILL.md — letter grade A+..F + Shields.io badge + SVG
• aif scan SKILL.md — OWASP AST10 security scan
• Pre-commit hook, GitHub Action, Claude Code plugin

Install

pip install aif-skillforge   # now works everywhere
aif check SKILL.md

GitHub Action

- uses: LiqunChen0606/skillforge@v0.6.2

Pre-commit

- repo: https://github.com/LiqunChen0606/skillforge
  rev: v0.6.2
  hooks:
    - id: aif-check

v0.6.0 — aif score (shareable badges)

New: grade your skill with a shareable badge

pip install aif-skillforge
aif score my-skill.md

Output:

SkillForge Score: my-skill.md
============================================================
  Score:  100/100  (A+)
  Lint:   0 errors, 0 warnings
  Security: 0 critical, 0 high, 0 medium, 0 low

Add a badge to your README

aif score my-skill.md --format shields -o badge.json
git add badge.json && git commit -m "add skillforge badge"

Then in your README:

![SkillForge](https://img.shields.io/endpoint?url=https://raw.githubusercontent.com/USER/REPO/main/badge.json)

Or use a self-contained SVG with no external dependency:

aif score my-skill.md --format svg -o badge.svg

Grade thresholds

Grade	Range	Meaning
A+	97-100	All checks clean
A / A-	90-96	One minor issue
B+ / B / B-	80-89	Several minor issues
C+ / C / C-	70-79	Multiple issues, review
D	60-69	Significant problems
F	<60	Critical security finding or parse failure

Scoring: 100 points start, lint errors -7, lint warnings -3, security Critical -15, High -8, Medium -3, Low -1.

CI gate

aif score skill.md --min-grade B    # exits 1 if below B

Also adds a pre-commit hook:

- id: aif-score
  args: [--min-grade, B]

And a /score-skill Claude Code plugin command.

Links

• PyPI: https://pypi.org/project/aif-skillforge/0.6.0/
• Full README: https://github.com/LiqunChen0606/skillforge#grade-your-skill-with-a-shareable-badge

v0.5.0 — Quality layer for SKILL.md

What's new in v0.5.0

SkillForge is now a pip-installable SKILL.md quality tool — no Rust build required for the common case.

Install

pip install aif-skillforge
aif check SKILL.md

Use it in three places

Pre-commit hook:

repos:
  - repo: https://github.com/LiqunChen0606/skillforge
    rev: v0.5.0
    hooks:
      - id: aif-check

GitHub Action:

- uses: LiqunChen0606/skillforge@v0.5.0
  with:
    path: ./skills

Claude Code plugin:

/plugin install LiqunChen0606/skillforge
/lint-skill my-skill.md

What it checks

• 10 structural lint checks: frontmatter, required sections, empty blocks, name format, hash verification, claim-evidence linkage, broken references, undefined terms
• 6 security checks (OWASP AST10 aligned): prompt injection, hidden Unicode, dangerous tools, external fetches, privilege escalation, data exfiltration

Breaking changes

• Legacy @end AIF syntax (v1) is no longer accepted. Use aif migrate-syntax <path> to convert existing .aif files.

Python API (16 functions)

parse, compile, render, lint, scan, infer, import_markdown, import_html, import_skill_md, export_skill_md, clean_html, hash_skill, generate_keypair, sign_skill, verify_skill, migrate_syntax

Links

• PyPI: https://pypi.org/project/aif-skillforge/0.5.0/
• Tutorial: tutorial/skillforge_tutorial.ipynb
• Document roundtrips: tutorial/skillforge_roundtrips.ipynb

v0.1.0 — SkillForge: Typed Semantic Documents for LLMs

SkillForge v0.1.0

Semantic document compiler and AI skill toolkit. Typed blocks improve LLM constraint resistance by +18pp vs raw Markdown.

Highlights

• 13 Rust crates — parser, 12+ output formats, skill toolkit, eval pipeline, migration engine, LSP server
• Python bindings — pip install skillforge via PyO3 (7 functions)
• Ed25519 skill signing — cryptographic tamper detection
• +18pp constraint resistance — typed tags help LLMs resist pressure to skip steps
• 7 benchmark suites — including adversarial constraint resistance testing
• 590+ tests across all crates

Install

```bash
cargo install --path crates/aif-cli
```

Key Results

Metric	Value
Constraint resistance (LML vs MD)	0.86 vs 0.68 (+18pp)
Hard scenarios	0.76 vs 0.65 (+11pp)
Token efficiency vs Markdown	22% fewer tokens, full semantic types
Case study (superpowers skills)	25% fewer tokens, 52 typed blocks