ecc

name	ecc
description	Everything Claude Code (ECC) — agent harness configuration layer for Claude Code and other AI coding IDEs. Augments agents with skills, hooks, persistent memory, model routing, and quality gates without modifying the underlying model.
license	MIT
compatibility	yana-ai >= 1.3.54
metadata	{"origin":"yana-ai — synthesized from affaan-m/ECC (MIT)","version":"1.0.0"}
triggers	["ECC","everything claude code","agent harness","agent harness optimization","claude code harness","coding agent harness","hook profile minimal strict","ECC_HOOK_PROFILE","agent skills hooks rules","autonomous loop quality gate","model route complexity","harness audit","loop-start loop-status"]
do_not_use_for	["Multi-agent pipelines with inter-agent communication — use crewai or smolagents","Building LLM apps or chatbots — use langgraph or vercel-ai-sdk","Fine-tuning or training models","Teams needing Python-native agent framework with eval tooling — use smolagents + deepeval"]
see_also	["smolagents","crewai","langgraph","agent-safety-patterns"]

ECC — Everything Claude Code Agent Harness

Source: affaan-m/ECC (MIT) — runtime configuration harness for AI coding agents

What ECC Is

A drop-in configuration layer for Claude Code (and Codex, Cursor, Gemini CLI, Zed, Copilot). It does not modify the model — it wraps the agent runtime with:

• Reusable skills (declarative SKILL.md docs auto-loaded into context)
• Hook automations (shell scripts on PreToolUse, PostToolUse, SessionStart, Stop)
• Rules (always-active guidelines injected every session)
• Memory (persistent facts across sessions)
• Model routing (send simple tasks to cheap model, complex to power model)

Directory Layout

.claude/
  skills/          ← declarative "when/how to use" docs
  hooks/           ← trigger-based shell scripts
  rules/           ← always-active session guidelines
  memory/          ← persistent facts
  settings.json    ← hook registration + model routing

Hook Profile Gating

# Control hook strictness without editing files
export ECC_HOOK_PROFILE=minimal    # only safety hooks
export ECC_HOOK_PROFILE=standard   # standard set (default)
export ECC_HOOK_PROFILE=strict     # all hooks, max enforcement

# In hook scripts — check profile:
[[ "${ECC_HOOK_PROFILE:-standard}" == "minimal" ]] && exit 0

Key Built-in Commands

/learn           — persist lesson to session memory after solving hard problem
/harness-audit   — audit active hooks + skills, detect drift
/loop-start      — begin autonomous multi-turn iteration loop
/loop-status     — check loop progress and quality gate results
/quality-gate    — enforce coding standards + test coverage + security before commit
/model-route     — route task to cheapest capable model by complexity score

Model Routing Pattern

// .claude/settings.json
{
  "modelRouting":