A small, nestable agent-loop job primitive: loop(body→until→review) composed with dag, pluggable engines, agent-judge gates, budget + run records. TypeScript.
@codeproxy/core is a zero-dependency TypeScript library that converts upstream LLM responses in Chat Completions or Anthropic Messages format into the OpenAI Responses API format, so that AI coding agents like Codex and Claude Code can work with any language model.
Harness-agnostic toolkit for AI coding agents: detect the harness (Claude, Codex, Cursor, Cline, Gemini, Aider...) and run a pure policy core. One Bun-native package.
Air-traffic control for AI agents editing a shared repo - semantic pre-flight collision detection over MCP. Model-agnostic (Claude Code, Cursor, Codex).
Never hit a rate-limit again. Route Claude Code, Cursor & 40+ providers through the subscription you already pay for, cut 40-80% tokens (RTK), stay un-banned. Self-hosted LiteLLM / OpenRouter / 9router alternative — one Go binary, anti-ban + sovereign P2P mesh.
Persistent-memory harness for AI coding agents — captures Claude Code/Codex sessions into an Obsidian vault, plus a native spec->change->verify loop. Local-first, zero runtime deps.
A coding-discipline spec for OpenAI's Codex CLI, enforced by native Codex hooks + a rule-hit telemetry loop. Turns an AGENTS.md the model drifts from into a system that holds. Independent of oh-my-codex.