arbiter-score 0.6.0

Deterministic, polyglot code quality scoring with governance integration

Arbiter

Agent-aware code quality scoring for multi-agent codebases.

In 2026, code is written by fleets of AI agents. Arbiter knows who wrote each line -- human or AI -- and scores quality accordingly.

Install

pip install arbiter                  # core (stdlib only)
pip install "arbiter[analyzers]"     # + ruff, radon, vulture, bandit

# Or from source
git clone https://github.com/hummbl-dev/arbiter.git && cd arbiter
pip install -e ".[analyzers]"

Usage

# Quick score -- no persistence, instant feedback
arbiter score /path/to/repo

# Full analysis with per-commit agent attribution
arbiter analyze /path/to/repo

# Agent leaderboard -- who writes the best code?
arbiter agents

# Start the dashboard (single HTML file, no build step)
arbiter serve --port 8080

What It Scores

Arbiter wraps tools you already trust and combines them into a deterministic composite score:

Analyzer	Tool	Weight	What It Finds
Lint	ruff	35%	Style violations, import errors, bugbear patterns
Security	bandit	30%	Hardcoded secrets, shell injection, dangerous patterns
Complexity	radon	35%	Cyclomatic complexity (grade A-F per function)
Dead Code	vulture	penalty	Unused functions, imports, variables
Duplication	AST hash	penalty	Near-duplicate function bodies
Semgrep	semgrep	opt-in	Custom rule enforcement (enable via config)

Scoring: 100 - (penalty / LOC) * normalization. Grades: A (90+) | B (80+) | C (70+) | D (60+) | F (<60).

What Makes Arbiter Different

Feature	Traditional Tools	Arbiter
Agent attribution	None	First-class: tracks Claude, Codex, Gemini, Copilot, humans
Per-commit scoring	Repo-wide only	Scores each commit's changed files individually
Diff analysis	N/A	Score only what changed in a PR/branch
Agent-specific gates	N/A	Different quality thresholds per agent trust tier
Dashboard	SaaS login	Single HTML file with per-agent timelines and fleet view
Dependencies	Heavy	Analysis tools only; core is stdlib Python

CLI Reference

arbiter analyze <repo>                     # Full analysis + per-commit scoring + persist
arbiter score <repo> [--json] [--exclude]  # Quick score (no persist)
arbiter diff <repo> [--base main] [--json] # Score changed files vs base branch
arbiter agents                             # Agent leaderboard
arbiter trend [--days 30]                  # Quality trend
arbiter worst [--limit 20]                 # Worst files
arbiter commits [--agent claude]           # Recent commits with scores
arbiter audit-fleet <directory>            # Audit all repos in a directory
arbiter triage                             # Auto-classify repos: green/yellow/red/archive
arbiter fix <repo> [--dry-run]             # Auto-fix ruff findings + before/after score
arbiter serve [--port 8080]                # API + dashboard

Tests

pip install ".[test]"
PYTHONPATH=src python -m pytest tests/ -v    # 78 tests, <7 seconds

Quality Gate

Arbiter grades itself on every push and PR. The CI runs arbiter score . and fails if the score drops below 90 (A grade).

# Run the same check locally
arbiter score . --fail-under 90

# Score only your changed files against main
arbiter diff . --base main --fail-under 80

Requirements

• Python 3.11+
• git (for historian)
• Optional: ruff, radon, vulture, bandit (install via [analyzers] extra)

HUMMBL Ecosystem

Part of the HUMMBL cognitive AI architecture:

• hummbl-governance -- Governance primitives that Arbiter scores against
• base120 -- 120 mental models for structured reasoning
• mcp-server -- MCP server for AI agent integration

Learn more at hummbl.io.

License

Apache 2.0 -- see LICENSE.

---

Built by HUMMBL LLC from production experience coordinating Claude, Codex, Gemini, and human engineers on a 14,000+ test codebase.