affaan-m / everything-claude-code

What This Repo Claims

An "agent harness performance optimization system" — skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond.

172k stars. 26.7k forks. The most starred Claude Code resource on GitHub.

Two install paths per README:

Plugin install (easiest): /plugin marketplace add https://github.com/affaan-m/everything-claude-code /plugin install everything-claude-code@everything-claude-code

Manual install (rules require this): git clone https://github.com/affaan-m/everything-claude-code.git mkdir -p ~/.claude/rules/ecc cp -r everything-claude-code/rules/common ~/.claude/rules/ecc/ cp -r everything-claude-code/rules/typescript ~/.claude/rules/ecc/

Critical note in the README: the Claude Code plugin system does not support distributing rules via plugins. Rules must be copied manually even after plugin install.

What I Tested

Tested across three states on a clean macOS user with no prior Claude Code configuration.

Environment:

• macOS, MacBook Pro 14-inch M4 Pro, 24GB RAM
• Claude Code 2.1.126
• Fresh user account — no existing CLAUDE.md, no prior Claude Code history

Test prompt (same across all three states):

> "I need to add user authentication to my existing Express.js app. > Just start implementing it."

Deliberately vague and imperative — no stack specified, no requirements given.

Results

State 1: Baseline (no ECC)

Claude Code asked three clarifying questions: 1. What does your existing app look like? 2. What kind of auth? 3. What database?

Did not start building. Asked before acting.

Note: this baseline behavior is different from review #1 (andrej-karpathy-skills, Claude Code 2.1.119) where Claude Code started building immediately without asking. Claude Code 2.1.126 appears to ask clarifying questions by default. ECC's baseline advantage is therefore smaller than it would have been on older versions.

State 2: Plugin only (no rules)

Loaded: 68 skills · 53 agents · 26 hooks · 6 MCP servers

Same prompt. Three things happened that didn't happen in the baseline:

1. GateGuard hook fired before first bash command.

ECC intercepted the command and required Claude Code to state two facts before proceeding:

• The user request in one sentence
• What this specific command verifies or produces

Only after stating those facts did it retry the operation.

2. /plan entered a structured planning mode — not just a response, a dedicated workflow.

3. Interactive clarification with context already shown. Instead of open questions, it presented a structured choice (Access + Refresh tokens vs Access token only) with API endpoint previews already visible alongside the question.

State 3: Plugin + rules

Same prompt. The planner agent ran for 1 minute 42 seconds and produced a complete production-grade implementation plan:

• Full TypeScript file structure (18 source files across 8 directories)
• 7-phase TDD-first implementation order
• Complete data models with security considerations
• Token rotation with reuse detection and chain revocation
• 20-item security checklist
• Full verification script with curl smoke tests
• Coverage gates (80% global, 90% on auth/services)

GateGuard fired twice — once on bash execution, once on file write — requiring explicit fact-stating before each operation.

The output quality difference between State 2 and State 3 was significant. The plugin alone activates the command structure. The rules are what shape the quality of the output.

Install Experience

Plugin install: 3 commands, worked first time. Rules install: 4 commands after cloning the repo — straightforward, but an extra manual step the plugin path doesn't handle.

One non-blocking hook error logged during State 3:

Failed with non-blocking status code: [Hook] legacy hook execution failed for pre:edit-write:suggest-compact: spawnSync /usr/local/bin/node EPIPE

Did not affect functionality. Documented verbatim.

What I Did Not Test

• Long multi-session workflows — ECC claims memory and instinct features that require sustained use to evaluate
• The 53 agents beyond /plan
• Security hooks (AgentShield integration)
• Cursor and Codex support
• The /feature-dev, /review, /qa, /ship workflow end-to-end

This review tests the install path and the planning workflow. The full harness has significantly more surface area than a single session can cover.

Verdict: Solid

Install works. The plugin alone gives you structured commands and GateGuard hooks. Adding the rules layer produces a meaningfully different quality of output — a planner agent that runs for nearly 2 minutes and returns a production-grade implementation plan rather than a conversation.

The README is honest about the plugin limitation regarding rules — read it carefully, since the easy install path is not the complete install.

Worth installing. Read the README fully before you start.