The Problem

Developers using Claude‑Code, Codex, or similar AI harnesses spend most of their time stitching together hooks, rules, and custom agents. The glue code proliferates, becomes hard to reason about, and performance degrades as session data grows. Maintaining a coherent, version‑controlled set of skills, instincts, and security checks is therefore a constant source of friction.

What This Does

The repository implements Everything Claude Code (ECC) – a self‑contained harness that bundles reusable skills, hooks, and rules for multiple AI agents. Core assets live under .opencode/ (TypeScript entry point index.ts) and scripts/ (the CLI driver claw.js). Hooks such as .cursor/hooks/adapter.js and .cursor/hooks/before-shell-execution.js intercept editor actions, while rule files in .cursor/rules/ encode coding‑style, security, and performance policies. The agents/ folder contains ready‑made agent prompts (e.g., agents/code-reviewer.md). Documentation (≈360 Markdown files) and test suites (62 tests) are shipped alongside the code.

How It Is Wired

Execution begins at scripts/claw.js (function main at line 278). main calls 57 functions across the codebase, most notably readFile, log, and a cascade of utility helpers from scripts/lib/utils.js.

  • scripts/lib/utils.js is the most‑connected module (27 functions, imported by six other files) and defines path helpers such as getClaudeDir and getSessionsDir. Because it is a hub, any change here can affect the entire CLI flow.
  • scripts/lib/session-aliases.js and scripts/lib/session-manager.js handle alias resolution and session persistence, each exposing five‑plus functions that read/write files (e.g., loadAliases, saveAliases).
  • Hook orchestration lives in .cursor/hooks/. The adapter (.cursor/hooks/adapter.js) is imported by 15 other modules, making it another blast‑radius hotspot. It routes editor events to specific hook implementations (before-shell-execution, after-file-edit, etc.).
  • The internal call graph shows 332 resolved function‑to‑function edges. The most frequently invoked internal utilities are readFile (11 callers) and log (11 callers).
  • Files that touch the outside world are limited to 31 functions that perform filesystem I/O. The shortest external path is main → readFile → fs.readFileSync. No network calls are evident in the static graph, so the runtime impact is confined to local file operations.

No circular dependencies appear, reducing the risk of hidden coupling. However, hub modules (utils.js, adapter.js) and several oversized test files (tests/ci/validators.test.js) represent the primary “change‑impact” zones.

How To Use It

# Clone the repo
git clone https://github.com/moses-y/everything-claude-code.git
cd everything-claude-code

# Install Node dependencies (npm is the lockfile format)
npm ci   # respects package-lock.json in the repo root and .opencode/

# Optional: install TypeScript for the .opencode package
npm install -g typescript   # needed only if you modify .opencode sources

# Run the CLI driver (the documented entry point)
node scripts/claw.js --help

Configuration files live under .opencode/ (opencode.json) and .cursor/ (hooks.json). The CLI reads those files to locate skill definitions (.agents/skills/…) and to load hook pipelines. No environment variables are required out‑of‑the‑box; the system operates on the local filesystem.

Real‑World Use

A development team integrates ECC into their CI pipeline to enforce coding standards on every pull request. The CI workflow invokes node scripts/claw.js run-tests which triggers the hook chain, validates rule compliance, and writes a session report under ~/.claude/sessions/. The report can be consumed by a downstream audit service without any network code changes.

Code Health & Issues

  • HIGH – Pin third‑party GitHub Actions to commit SHAs (.github/workflows/*).
  • HIGH – Discarded exit codes mask failures (maintenance.yml line 23).
  • HIGH – continue-on-error on correctness steps hides test failures (ci.yml line 195).
  • MEDIUM – No Dependabot/Renovate config (.github/dependabot.yml missing).
  • MEDIUM – Large binary (assets/images/longform/01-header.png, 5.4 MiB) should be moved to Git LFS.
  • MEDIUM – Checkout step keeps token; set persist-credentials: false (ci.yml).
  • MEDIUM – postinstall script fetches remote resources; consider moving to explicit build step (package.json).
  • LOW – Add timeout-minutes to jobs in maintenance.yml to avoid runaway runs.

The repository includes a full test suite, CI definitions, a LICENSE, and lockfiles, indicating reasonable hygiene. Duplicate hook implementations and oversized test files (e.g., tests/ci/validators.test.js with 1,669 lines) are flagged by the static analysis and should be refactored for maintainability.

The Bottom Line

ECC provides a concrete, file‑based framework for stitching AI‑agent hooks, skills, and rules together, with clear entry points and modest external side effects. Its main drawbacks are a few high‑impact hub modules and some maintenance‑level technical debt (pinned actions, large binaries). Teams needing a configurable, locally‑runnable harness for Claude‑Code or Codex will find a solid foundation, but should allocate effort to reduce hub complexity and clean up duplicated hook logic before extensive customization.