Skip to content

Proxie

Spectra deploys 8 Claude Opus 4.6 agents — 6 parallel specialists, a MetaPrompter, and a CritiqueAgent with Extended Thinking — to analyze entire codebases in under 5 minutes.

Code Video Cerebral Valley

Core trick

A planner agent spawns six parallel specialist agents over a repository, then a critique agent validates and merges their findings.

Tags

dev toolssecurity

Description

Spectra deploys 8 Claude Opus 4.6 agents — 6 parallel specialists, a MetaPrompter, and a CritiqueAgent with Extended Thinking — to analyze entire codebases in under 5 minutes. It replaces weeks of manual code review for VCs, CTOs, and engineering leaders evaluating unfamiliar repositories. Given a GitHub URL, Spectra ingests the codebase, plans an analysis strategy (MetaPrompter), launches 6 specialist agents in parallel across architecture, security, quality, documentation, dependencies, and performance. Results are merged, deduplicated, and validated by a CritiqueAgent using Extended Thinking — producing a scored report with letter grades, code-level recommendations, and compliance readiness assessments (OWASP Top 10, SOC 2, PCI DSS 4.0, NIST CSF). Analyzed Anthropic's own SDK (576 files) in 164 seconds for $15.19, producing 56 findings with 0 critical issues and an overall score of B (83/100).
leocder07/spectra★ 190 forksPythonMITCreated Feb 14, 2026 · last push Jun 13, 2026

README excerpt

<div align="center">

# SPECTRA

### The full spectrum of your codebase

**Spectra grades any repository on architecture, security, quality, docs, maintainability, and performance — in minutes.** It runs 8 specialized Claude agents (6 specialists in parallel, plus a planner and a critic) on Opus 4.7, so you get a full audit instead of a single linter's opinion. Built for developers running self-checks, teams gating PRs in CI, and reviewers who need a second pair of eyes before merge.

#### <a id="disclaimer"></a>Disclaimer

**Indicative analysis — not auditor-grade evidence.** Spectra runs 8 LLM agents over your code; findings are heuristic and require human verification before being treated as compliance evidence, audit input, or pass/fail signal in regulated workflows. Use Spectra as a fast directional signal — pair it with deterministic SAST/DAST tooling and a human reviewer for anything compliance-bound. The same notice ships in every HTML, JSON, and SARIF report ([source](src/spectra/entities/disclaimer.py)).

[![Python 3.12+](https://img.shields.io/badge/python-3.12+-7C3AED?style=for-the-badge&logo=python&logoColor=white)](https://python.org)
[![Tests](https://img.shields.io/badge/tests-2%2C633_passed-22C55E?style=for-the-badge)](tests/)
[![Coverage](https://img.shields.io/badge/coverage-97%25-22C55E?style=for-the-badge)](tests/)
[![License: MIT](https://img.shields.io/badge/license-MIT-F59E0B?style=for-the-badge)](LICENSE)
[![Built with Claude](https://img.shields.io/b

Team