A drop-in harness that makes Claude Code cheaper, sharper, and forgetful-proof: persistent memory, token savings, anti-drift, a code-map RAG, disciplined cheap-model coding.
Agent Claude Code
Read-only AUDITOR, Codex voice (codex exec --sandbox read-only). Use to audit a Claude-built diff or unit against its CONTEXT/CHANGE/GOAL/VERIFY spec; reports findings and stops. Never edits or spawns; never fixes.
Agent Claude Code
The BUILDER: sends ONE unit to the synchronous Codex CLI (codex exec), which edits files, and returns the diff. Use for a decomposed unit or a fix with a known target. Not for planning, audits, research. Not codex:codex-rescue (a forwarder).
Agent Claude Code
Real Gemini 2.5 (Pro/Flash) via the gemini shim to OpenCode, with its own Read/Bash/Edit tools. Use for large-context passes, image/video vision, scraping and data pulls, a cheap second opinion, or the non-Claude council voice.
Agent Claude Code needs its repo
After a unit is built and verified, before commit: regenerate cached maps (.codemap, graphify), verify live ones (L0/L1/skeleton). Use when files are added, deleted, moved or renamed; not for in-function edits. Never edits source.
Agent Claude Code
Read-only AUDITOR (Claude). Use to audit a Codex-built diff or unit against its CONTEXT/CHANGE/GOAL/VERIFY spec, or as the Claude voice in a 2-model council. Reports findings and stops; never edits or spawns; fixes go to the builder.
Agent Claude Code
Read-only AUDITOR pinned to Claude Opus 4.8 (Ro 2026-09-22: "Opus 5.5 para construir y Opus 4.8 para auditar"). Reads a finished unit against its CONTEXT/CHANGE/GOAL/VERIFY spec, runs its tests, reports findings by severity, and stops. Never edits, never spawns.
At most 3 mods per repository are shown here, and a mod shipped inside a plugin is left to that plugin's page — the rest are on their repository pages: