// the find
affaan-m/ECC
The agent harness performance optimization system. Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond.
ECC is a collection of prompts, rules, hooks, and slash-command definitions meant to make AI coding agents (Claude Code, Cursor, Codex, etc.) more consistent and productive. It's aimed at developers who spend serious time in AI-assisted workflows and want a shared, versioned system for that instead of one-off CLAUDE.md files.
The cross-harness scope is real — the same skills and rules ship for Claude Code, Cursor, OpenCode, and Codex, with per-harness adapters rather than just copying the same markdown everywhere. The hook system for session memory and continuous learning (capturing patterns at Stop time and injecting them at SessionStart) is one of the more thoughtful things here; most people do this ad hoc. The selective install architecture with a manifest-driven plan/apply split means you don't have to take the whole thing. Language coverage is broad: TypeScript, Go, Python, Java, Kotlin, Rust, C++, Swift, PHP, Perl — most are first-class, not afterthoughts.
227K stars on a repo that is essentially a prompt library is a stat to hold lightly — the star count looks inflated relative to any evidence of real adoption depth. The install story is genuinely confusing: plugin path vs. manual path vs. npm path, with explicit warnings not to combine them, an uninstall script because people keep stacking them, and a repair command for when things go wrong. That's a sign the install surface has grown past what the architecture can cleanly support. The 'operator' and business workflow additions (brand-voice, investor-outreach, social-graph-ranker) feel like scope drift from the core use case and are unlikely to be what most developers are here for. The Rust control-plane prototype in ecc2/ is shipped as an alpha with no clear graduation path, which means it'll sit there and confuse people for a long time.