ECC (Everything Claude Code): What It Is and Whether It's Worth Using

1. Don’t let Star count decide: what pain does ECC actually solve?

Most people install ECC because “everyone is.” A better question: does your Claude Code / Cursor lack IQ, or lack process? If you re-explain the project every session, merge without confidence, or Hooks spam the terminal, ECC has a role. If you only ask two-line script questions, a handful of project Rules is enough.

affaan-m/ECC calls itself an agent harness performance optimization system — it serves the harness (Claude Code plugin, Cursor Skills, Codex AGENTS.md), not another chat API. Author Affaan Mustafa open-sourced ~10 months of production config after an Anthropic Hackathon win. Counts below match your clone’s README:

• Skills (~246): SKILL.md workflow units — passive triggers, agent reuse; preferred for new features;
• Agents (~61): role sub-agents for review, build fix, multilingual review;
• Hooks: SessionStart/Stop scripts for cross-session memory, summaries, runtime gates;
• Rules per language (TS/Python/Go/Java…), install only what you need;
• MCP configs, Commands (legacy slash shims in legacy-command-shims/), AgentShield;
• v2.0.0-rc.1: Hermes operator narrative + ecc2/ Rust control plane (alpha, not general release).

Read README.zh-CN.md for Chinese overview; pair with the author’s Shorthand Guide and Longform Guide (token optimization, parallel worktrees, eval loops). Core repo is MIT; ECC Pro (GitHub App, private repo audit) is optional hosted service.

2. Core modules: Skills, Hooks, memory, and security

2.1 Skills: workflows as first-class citizens

ECC treats Skills as the primary surface; Commands are maintenance/compat. Customize teams by forking skills/, not stacking slash commands. Coverage spans search-first, eval-harness, Django/Spring patterns, and v2 media/ops packs. For individuals, 5–10 stack-matched Skills beat --profile full.

2.2 Hooks and memory: sessions that don’t “forget”

The Longform Guide highlights Hooks that summarize on session end and restore on SessionStart, cutting re-pasted PRD tokens. v1.8+ adds ECC_HOOK_PROFILE=minimal|standard|strict and ECC_DISABLED_HOOKS to tame noisy Hooks without editing files — critical when “every save runs two scripts.”

2.3 AgentShield and research-first culture

Security: /security-scan, AgentShield (Security Guide, Agentic Security post). Culture: research before implement — check official docs and run verification, don’t guess APIs. Same spirit as our rent-a-Mac onboarding checklist: reproducible build before automation.

3. Cross-harness: Claude Code, Cursor, Codex — one pack?

README claims Claude Code, Codex, Cursor, OpenCode, Gemini CLI, Zed, Copilot. Repo has .cursor/, .codex/, .claude/ — one Skills/Rules set mapped to multiple IDEs, not unrelated prompt dumps per tool.

v1.8 commands like /harness-audit, /quality-gate, /model-route productize “is my harness healthy?” — run after two weeks, not day one.

4. Install pitfalls: why “it got slower after install”

Quick Start says in bold: pick exactly one install path. Real failures: marketplace install Monday, colleague says “full is complete,” install.sh --profile full Wednesday — duplicate /eval, Hooks fire twice, thousands of mystery tokens. That’s stacking, not ECC being slow.

Recommended (Claude Code):

1. Plugin (marketplace) + copy only needed language rules; or
2. Manual: ./install.sh --profile minimal --target claude; add --modules hooks-runtime when you need gates.
3. Preview with npx ecc consult "security reviews" --target claude before committing.

Windows: install.ps1; Hooks largely Node.js since v1.7 (fewer bash-only failures). Messed up? Reset/Uninstall per README — Discussions for community threads. See TROUBLESHOOTING and CONTRIBUTING.

5. Worth it? Match your role

ECC is worth trying as an agent engineering base — not worth blind full install without docs. Stars mean “someone open-sourced extreme Claude Code craft,” not “3× speed on clone.”

6. Cloud Mac: agents on the laptop, builds in the datacenter

ECC handles IDE/CLI reasoning and workflow; xcodebuild, simulators, notarization, Keychain stay on real macOS. Common split:

• Local/Cursor + ECC for task breakdown, patches, PRs;
• Cloud Mac mini M4 for CI, parallel worktrees, long launchd jobs (remote Mac scheduled agent FAQ);
• Release week: daily lease to validate 16GB/24GB and agent peaks, then weekly/monthly.

Closing the laptop lid doesn’t run Xcode in the cloud — that’s why harness and build machine should decouple. Compare cloud Mac procurement guide for lease tiers.

7. FAQ

Conflict with Cursor Rules? Possible overlap. Pick “ECC primary” or “project Rules primary”; diff .cursor/ before merge; try /harness-audit.

Must I pay for ECC Pro? No — OSS MIT. Pro is GitHub App / private audit hosting.

What are Hermes / ecc2? rc.1 narrative + Rust ecc2/ alpha; production stays Plugin/manual per README.

How to track updates? Watch Releases; preview with install-plan (v1.9+) before major bumps.

Same as OpenClaw? No — OpenClaw is gateway execution; ECC is coding harness performance and engineering norms. Layer them; don’t fight for ports, MCP, and Hooks on one host.

8. References (external)

• Repo: github.com/affaan-m/ECC
• Chinese README: README.zh-CN.md
• Contribute & security: CONTRIBUTING · SECURITY · TROUBLESHOOTING
• Community: Discussions
• Author guides (X): Shorthand · Longform

9. Closing

ECC moves you from “using Claude Code” to governing an agent harness — Skills workflows, Hook memory, auditable install and security scans. For individuals, minimal + curated Skills buys a research-first habit; for teams, a two-week PoC beats Star-driven full install. Running iOS/macOS pipelines? Keep the coding harness in the IDE and the build environment on a cloud Mac — daily lease until xcodebuild is green, then talk full agent hosting.