hummbl-dev/arbiter

Arbiter

Agent-aware code quality system for multi-agent codebases. Wraps ruff, Bandit, radon, and vulture with per-agent attribution and transparent decomposed scoring.

A

98.0

Overall Quality Score
Audited: 2026-03-25 Version: v0.2.0 Python 3.11+
89 Tests
12 Modules
6 Analyzers
0 High Findings
Apache-2.0 License

Category Breakdown

100
Lint
A
100
Security
A
96
Complexity
A
95
Dead Code
A
99
Duplication
A

Key Modules

Module Purpose Status
api.py HTTP server with JSON endpoints for scoring, agents, trends Pass
scoring.py Quality scoring logic with weighted category aggregation Pass
store.py Data persistence for audit results and history Pass
git_historian.py Git commit attribution and per-agent history analysis Pass
diff_analyzer.py Diff-based scoring for changed files only Pass
report.py HTML/PDF audit report generation Pass
agent_registry.py Agent identity and trust tier configuration Pass
bus_bridge.py Coordination bus integration Pass
analyzers/ Pluggable analysis tools: ruff, Bandit, radon, vulture, duplication Pass

Findings

Low Minor complexity in store.py

The store.py module (10KB) has slightly elevated cyclomatic complexity in the query filtering path. Consider extracting filter predicates into separate functions for testability.

Info Optional dependencies well-isolated

Runtime analysis tools (ruff, radon, vulture, Bandit) are wrapped in try/except imports with clear error messages. The core scoring logic works without them installed.

Want an audit like this for your codebase? Free for open-source repos.

Request Free Audit