Plugin

eight-eyes

Multi-agent code review with 8 hook-enforced roles. Each role targets a different failure surface — the skeptic reviews blind without the implementer's narrative, the security auditor can only run approved scan commands, the test writer cannot touch source code, and the implementer cannot run Bash. Scope boundaries are enforced through deterministic Python hooks at the PreToolUse layer, not through prompt instructions. The 8 roles: implementer (scoped writes, no Bash), test-writer (test directories only), skeptic (blind read-only review), security auditor (read-only + approved scan commands), performance profiler (benchmarks in isolated worktree), accessibility checker (a11y tools + WCAG 2.1 checks), docs writer (doc paths only), and verifier (approved verification commands only). Features include parallel audit phase (4 review agents run concurrently), TDD mode (hook-level test-first enforcement), custom roles via manifest (--custom-role), REVIEW.md injection for project-specific review criteria, mission timeout and failure tracking. 96 tests. 100% Python stdlib. Zero external dependencies.

• Published source
• Repository

Analysed

11 October 2026

Last observed

12 October 2026

What the code does

• Runs shell commands Code that starts system commands on the machine it runs on.

Having a capability is not wrongdoing; many tools need these to do their job. It tells you what to check before you install.

This is public metadata, not a safety certification or complete assessment. Static analysis cannot establish what a tool does at runtime.