SkillHub

repo-analyzer

v1.2.0

GitHub repository trust scoring and due diligence. Use when asked to analyze, audit, score, or evaluate any GitHub repo — especially for crypto/DeFi project DD, checking if a repo is legit, evaluating code quality, verifying team credibility, or comparing multiple repos. Also handles X/Twitter URLs...

Sourced from ClawHub, Authored by Don-GBot

Installation

Please help me install the skill `repo-analyzer` from SkillHub official store. npx skills add Don-GBot/repo-analyzer

Repo Analyzer

Zero-dependency GitHub trust scorer. Runs 29 analysis modules across 12 scoring categories.

Usage

# Single repo
node scripts/analyze.js <owner/repo or github-url> [flags]

# From a tweet (auto-extracts GitHub links)
node scripts/analyze.js <x.com-or-twitter.com-url> [flags]

# Batch mode
node scripts/analyze.js --file <repos.txt> [--json]

Flags

  • --json — JSON output (for pipelines)
  • --oneline — compact one-line score
  • --badge — shields.io markdown badge
  • --verbose — show progress
  • --token <pat> — GitHub PAT (or set GITHUB_TOKEN env)
  • --file <path> — batch mode, one repo per line (# comments ok)

Environment

CRITICAL: Always run with GITHUB_TOKEN loaded. Without it, scores are severely degraded (missing stars, forks, commits). Before running: source ~/.bashrc (token is in ~/.bashrc as GITHUB_TOKEN). Or pass explicitly: GITHUB_TOKEN="$(grep GITHUB_TOKEN ~/.bashrc | cut -d'"' -f2)" node scripts/analyze.js ...

Scoring (14 categories, 168pts normalized to 100)

Category Max What it checks
Commit Health 20 Human vs bot, GPG sigs, code dumps, fake timestamps
Contributors 15 Bus factor, contributor diversity
Code Quality 25 Tests, CI, license, docs, lock files
AI Authenticity 15 AI slop detection in code/README
Social 10 Stars, forks, star/fork ratio, botted stars
Activity 10 Recent pushes, releases
Crypto Safety 5 Token mints, rug patterns, wallet addresses
Dependency Audit 10 Known malicious packages, typosquatting, install hooks, lock files
Fork Quality 8 Fork divergence, suspicious changes, gutted vs meaningful forks
README Quality 10 Install guide, examples, structure, API docs
Maintainability 10 File sizes, nesting, code/doc ratio
Project Health 10 Abandoned detection, velocity, issue response, PR review
Originality 5 Copy-paste, template detection, backer verification
Agent Safety 15 Install hooks, prompt injection, secrets, CI audit, permissions

Grade Scale

  • A (85+): LEGIT
  • B (70-84): SOLID
  • C (55-69): MIXED
  • D (40-54): SKETCHY
  • F (<40): AVOID

Key Features

  • Enhanced dependency audit: Detects known malicious packages (event-stream, ua-parser-js, etc.), typosquatting attacks, install hooks, and estimates transitive dependency bloat
  • Fork comparison: Analyzes fork divergence, detects cosmetic vs meaningful changes, flags suspicious modifications (removed CI, added wallets), identifies gutted forks
  • Agent safety: Detects prompt injection, credential harvesting, install script hooks, obfuscated code
  • Secrets detection: Finds hardcoded API keys, tokens, private keys via regex + entropy
  • Network mapping: Categorizes all outbound domains (API, CDN, unknown)
  • CI/CD audit: Checks GitHub Actions for pull_request_target, unpinned actions, secret leaks
  • Permissions manifest: Summarizes what the code needs to run (like an app permissions list)
  • Author reputation: Org memberships, suspicious repos, account age
  • Backer verification: Cross-references investor claims vs committer org membership
  • Complexity hotspots: Flags large files with deep nesting and high conditional density

Batch File Format

# One repo per line, # for comments
Uniswap/v3-core
https://github.com/aave/aave-v3-core
OpenZeppelin/openzeppelin-contracts

Output

Default: rich terminal report with bar charts, sections, verdict. --json: Full structured data for programmatic use. --oneline: RepoName: 85/100 [A] — 2 flags

When Reporting to User

Keep it concise. Lead with score/grade and notable findings. Skip sections with nothing interesting. Example:

"Uniswap/v3-core scored 75/B — 96% GPG-signed, 11 authors, MIT license. Flagged: abandoned (466 days no push), 2,597 transitive deps (bloated), secrets in CI run commands. Agent safety: CAUTION."