autoforge

The only OSS toolchain that measures, recommends, and fixes repository health in a single deterministic invocation — across Python, JavaScript/TypeScript, Go, and Rust.

One command to improve any repo

pip install -e .
autoforge-improve --dir /path/to/any-repo --apply --pr-summary PR.md

This runs the full closed loop:

score  →  suggest  →  fix  →  re-score  →  PR summary

No LLM. No API key. Deterministic and reproducible.

Proven on Django (80K stars)

Before:  51% D
After:   72% C  (+21 points, 487 fixes applied, ~45 seconds)

Full case study with before/after breakdowns and reproduction instructions.

What makes autoforge unique

Capability	autoforge	SonarQube	CodeClimate	Snyk	ruff/ESLint
Measure (whole-repo rubric)	6 dimensions	Issues only	Grades	Security only	Per-file
Recommend (file:line targets)	Ranked by impact	Issues	Grades	Advisories	Warnings
Fix (deterministic auto-apply)	AST/regex + templates	---	---	Deps only	Per-rule
Orchestrate (one-command loop)	autoforge-improve	---	---	---	---
Polyglot	Py + JS/TS + Go + Rust	Paid SaaS	Paid SaaS	---	Single-lang
OSS	MIT	Partial	---	---	Per-tool

The 6-dimension health rubric

Dimension	What it measures
Docs	README, CONTRIBUTING, SECURITY, LICENSE, AGENTS.md (md/rst/adoc)
Testing	Test framework detection + file count tiers (1/5/20/50)
CI/CD	GitHub Actions, GitLab CI, Jenkins, CircleCI, Travis
Security	SECURITY policy + dependency manifest for audit
Quality	Lint config, test/source ratio, file size, comment-marker debt
AI-Readiness	Agent instructions + type-hint coverage + docstring coverage + examples dir

AI-Readiness is unique to autoforge. No other health scorer measures how prepared a repo is for AI coding assistants.

Public leaderboard (8 major Python OSS, ~197K combined stars)

Rank	Repository	Stars	Total	Grade
1	autoforge	new	100%	A
2	matplotlib	20K	77%	B
3	astropy	4.5K	70%	C
4	sphinx	6.7K	62%	C
5	pytest	12K	59%	D
6	scikit-learn	61K	59%	D
7	sympy	13K	59%	D
8	django	80K	51%	D

Full leaderboard with AI-Readiness column

CLI tools (11 entry points)

Command	Purpose
autoforge-improve	Full closed loop: score + suggest + fix + re-score + PR summary
autoforge-score	6-dimension health score with badge/JSON output
autoforge-suggest	Prescriptive improvement plan (file:line targets, ranked by impact)
autoforge-fix	Deterministic auto-fix (docstrings, AGENTS.md, SECURITY.md, lint config, ...)
autoforge-leaderboard	Multi-repo ranked health comparison
autoforge-compare	Side-by-side score comparison of 2+ repos
autoforge-history	Score trend chart over time
autoforge-init	Analyze any repo and propose improvement goals
autoforge-tools	SWE-bench harness tools (search, view, edit, run-cmd)
autoforge-solve	SWE-bench task resolver
autoforge	Full SDLC pipeline (Survey + Idea + Implement + Test + Security + Stage + Approval)

What autoforge-fix auto-applies

Fixer	Language	Action
AGENTS.md	All (language-aware)	Creates a stub agent briefing with correct build/test commands
CONTRIBUTING.md	All	Creates a contributor guide stub
SECURITY.md	All	Creates a vulnerability reporting policy
examples/	All	Creates examples/hello.py
Lint config	Python	Appends [tool.ruff] to pyproject.toml
Docstrings	Python	AST-driven """docstring""" insertion
JSDoc	JS / TS	Regex-driven /** comment */ insertion
godoc	Go	// comment insertion above each func
rustdoc	Rust	/// comment insertion above each fn

All fixers are idempotent, process functions in descending line order (so earlier insertions don't corrupt later ones), and use a shared _docstring_text_for(name) to produce the same phrasing across all four languages.

Architecture

SDLC pipeline (7 agents)

Survey → Idea (tree search) → Implement → Test → Security (OWASP) → Stage → Approval

SWE-bench resolver stack

Component	Based on
Multi-stage Localizer	MASAI (75% localization rate)
Reproducer	SWE-agent fail-to-pass pattern
ACI Toolkit	SWE-agent Agent-Computer Interface
Search Subagent	WarpGrep (isolated context)
Interactive Resolver	ReAct tool-use loop

Harness mode (no API key)

autoforge runs within Claude Code's subscription — no ANTHROPIC_API_KEY needed.

Development

# Install
pip install -e .

# Run tests (494 tests, ~20s)
PYTHONPATH=. python3 -m pytest tests/ -v

# Score this repo
autoforge-score --dir .

# See what to improve
autoforge-suggest --dir . --top 5

# Auto-fix (dry-run by default)
autoforge-fix --dir . --apply

License

MIT