A command-line tool for analyzing Arch Linux AUR PKGBUILD files to identify security risks. Uses static analysis with optional AI-powered inspection.
aur_checker inspects PKGBUILD files for security vulnerabilities and suspicious patterns before you build packages. It combines context-aware static analysis with optional AI review to produce explainable risk assessments.
Analysis pipeline:
PKGBUILD → Fetch → Static Analysis → Metadata → Trust Signals → Scoring → [AI Review] → Risk Verdict
Key features:
- ✓ Context-aware detection to reduce false positives
- ✓ Explainable risk scoring with detailed breakdowns
- ✓ AUR trust signals (maintainer history, popularity, age)
- ✓ Optional AI analysis (works without API key)
- ✓ User-friendly output with plain language summaries
- ✓ JSON output with stable schema for CI integration
- ✓ Clear exit codes for automation
git clone https://github.com/programmersd21/aur_checker.git
cd aur_checker
pip install -e .Requires Python 3.10 or later.
# Check a single package
aur_checker check keepassx2
# Output as JSON
aur_checker check --json keepassx2
# Verbose mode with detailed signals
aur_checker check --verbose keepassx2
# Check multiple packages
aur_checker batch keepassx2 visual-studio-code-bin
# Check from a file
aur_checker batch --file packages.txt
# AI analysis from previous results
aur_checker explain --input analysis.json
# Build with makepkg
aur_checker install keepassx2
# Clear cached analysis
aur_checker clear-cacheAI analysis is optional. aur_checker works without an API key using static analysis only.
To enable AI analysis, set your Google Generative AI API key:
# Linux/macOS
export AURCHECKER_AI_API_KEY="your-api-key"
# Windows PowerShell
$env:AURCHECKER_AI_API_KEY="your-api-key"| Variable | Purpose | Default |
|---|---|---|
AURCHECKER_AI_MODEL |
AI model to use | gemini-3.1-flash-lite |
AURCHECKER_AI_TIMEOUT |
Request timeout (milliseconds) | 120000 |
- Fetch - Downloads PKGBUILD from AUR CGit
- Static Analysis - Context-aware pattern detection (skips comments, distinguishes legitimate operations)
- Metadata - Retrieves AUR RPC v5 package information (maintainer, votes, age)
- Trust Signals - Evaluates maintainer history, popularity, update patterns, out-of-date status
- Scoring - Computes 0-100 risk score with explainable breakdown of contributing factors
- AI Review (optional) - Heuristic analysis for additional context if API key is set
- Verdict - Final assessment: Safe to review / Needs attention / High risk
| Score | Risk Level | Recommendation | Exit Code |
|---|---|---|---|
| 0-20 | Low | Safe to review | 0 |
| 21-50 | Medium | Manual review recommended | 2 |
| 51-100 | High | Do not use | 1 |
Note: Without AI, static score is used directly (more conservative). With AI enabled, final score is 70% static + 30% AI.
Human output (default):
- User Summary: Simple verdict, key findings in plain language
- Technical Details: Score breakdown, trust signals, metadata
- AI Analysis: Heuristic insights (clearly labeled, if enabled)
- Verbose mode: All detected signals with
--verbose
JSON output (--json):
- Stable schema (v1.0.1) for automation
- Includes score_breakdown, trust_signals, analysis_mode
- Use with CI tools and security pipelines
0 - Analysis completed, package appears safe (ALLOW)
1 - Analysis completed, high risk detected (DENY)
2 - Analysis completed, manual review needed (REVIEW)
3 - Package not found on AUR
4 - Unrecoverable error during analysis
The static analyzer detects:
| Signal | Pattern | Risk Weight | Rationale |
|---|---|---|---|
remote_exec |
Downloads piped to shell (curl/wget → bash) | 50-70 | Direct code execution from internet |
obfuscation |
Base64, hex, eval, encoding layers | 15-35 | Attempts to hide behavior |
system_mod |
Writes to /etc, /boot, /usr/lib/systemd | 30 | Modifies sensitive system locations |
maintainer_changed |
Maintainer change detected | 15 | Increased risk during transitions |
orphan_adopted |
Package without active maintainer | 10 | No owner to respond to issues |
pkg_manager |
npm, pip, cargo, go, gem (outside deps) | 8-12 | Risk if used in build scripts |
external_calls |
Non-source HTTP/HTTPS URLs in execution | 5-15 | Network activity in build context |
| Signal | Description |
|---|---|
| Package age | Days since first submission |
| Maintainer history | stable / changed / orphan_adopted / new |
| Update frequency | Days since last update |
| Popularity | Vote count, >100 votes = popular |
| Out of date | Flagged by users as outdated |
✅ CAN detect:
- Direct remote code execution patterns (
curl | bash) - System modification to sensitive directories
- Encoding/obfuscation indicators
- Suspicious trust signals (orphaned, new maintainer, out-of-date)
- Risky patterns in build scripts
❌ CANNOT detect:
- Sophisticated multi-stage obfuscation
- Logic bombs or conditional malware
- Compromised upstream sources (even with checksums)
- Post-install behavior of compiled binaries
- Typosquatting or social engineering
See SECURITY.md for the complete threat model, limitations, and recommendations.
# Unit tests
pytest
# Type checking
mypy .
# Code style
ruff check .aur_checker/
ai/ # AI integration (optional)
cache/ # Result caching
cli/ # Command-line interface
core/ # Pipeline and context
features/ # Static analysis (context-aware)
output/ # Human and JSON formatters
scanner/ # PKGBUILD and metadata fetching
scoring/ # Risk scoring with breakdown
compat_platform/ # Cross-platform compatibility
- AI analysis requires valid API key and internet connection
- Static analysis uses context-aware patterns but may miss sophisticated obfuscation
- Risk scores are heuristic-based and should inform, not replace, manual review
- AI responses are non-deterministic and clearly labeled as heuristic
- Cannot analyze post-build behavior or compiled binary contents
Always read PKGBUILDs manually before building AUR packages.
Contributions welcome. Please open an issue or pull request on GitHub.
MIT License. See LICENSE.md for details.
See SECURITY.md for the threat model, what aur_checker can and cannot detect, and how to report security issues.
Maintainer: geniussantu1983@gmail.com
Repository: github.com/programmersd21/aur_checker