Cross-harness performance optimization system for AI coding agents, featuring 38+ agents, 183+ skills, multi-language rules, and integrated security auditing.
Everything Claude Code (ECC) is a systematic performance optimization suite for AI coding agents. According to the project's own README, it originated from an Anthropic hackathon award-winning project (note: this claim has not been independently verified with an official Anthropic announcement), iterated over 10+ months of intensive daily use to v1.10.0. Rather than a simple configuration collection, it is a full-lifecycle optimization system covering agent session lifecycles, comprising 38+ specialized agent definitions, 183+ reusable skill modules, 79+ backward-compatible command shims, 12+ language-specific coding rule sets, session lifecycle hooks, and MCP server configurations.
Core Capabilities#
Token & Context Optimization#
- Model selection routing based on task complexity
- System prompt compression for token savings
- Background process management to reduce context pollution
- Memory Persistence via cross-session automatic context save/load Hooks
- Continuous Learning: automatic pattern extraction from sessions into reusable skills
- NanoClaw v2 runtime: session branching, search, export, compression, and metrics collection
Verification & Quality Assurance#
- Verification loops with checkpoint evaluation, continuous evaluation, and pass@k metrics
- Multi-language TDD-driven development
- Enforced quality gates via
/quality-gatecommand - 997+ test cases
Security#
- AgentShield integration: 1,282 test cases + 102 security rules
- CVE tracking and one-click security scanning (
/security-scan)
Parallelization & Orchestration#
- Git worktrees parallel development, cascade task decomposition
- PM2 multi-instance scaling and subagent orchestration
- Multi-agent collaboration commands (
/multi-plan,/multi-execute) with ccg-workflow runtime
Cross-Harness Adaptation#
ECC's core differentiator is cross-harness unified adaptation, providing consistent workflow experiences across 8+ mainstream tools through dedicated configuration directories:
| Harness | Config Directory |
|---|---|
| Claude Code | .claude/ + .claude-plugin/ |
| OpenAI Codex | .codex/ + .codex-plugin/ |
| Cursor | .cursor/ |
| OpenCode | .opencode/ |
| Google Gemini | .gemini/ |
| AWS Kiro | .kiro/ |
| CodeBuddy | .codebuddy/ |
| Trae | .trae/ |
Installation & Usage#
Three installation paths: Claude Marketplace plugin (recommended), manual OSS clone, and npm (npx ecc-install --profile full). Manifest-driven selective installation with SQLite state tracking across Shell/PowerShell/npm.
Post-installation requires manual Rules copy to ~/.claude/rules/ (Claude Code plugin cannot auto-distribute rules).
CLI Tools#
ecc list-installed— List installed componentsecc doctor— Diagnose installation statusecc repair— Fix installation issues
Runtime Configuration#
ECC_HOOK_PROFILE=minimal|standard|strict— Control hook strictness levelECC_DISABLED_HOOKS=...— Disable specific hooks at runtime
Architecture Highlights#
- Build tools: Node.js (ESLint, Prettier, Commitlint), Python (Dashboard GUI via Tkinter), Rust (ECC 2.0 control plane, alpha)
- Config formats: YAML (agent.yaml), JSON (.mcp.json, manifests), Markdown (skills/rules)
- CI/CD: GitHub Actions
- Protocol integration: Model Context Protocol (MCP) via
mcp-configs/directory - ECC 2.0: Rust control plane prototype (alpha) with dashboard, start, sessions, status, stop, resume, daemon commands; GA timeline TBD
Unverified Information#
- Anthropic hackathon award lacks official corroboration link
- ECC Tools GitHub App pricing details not found in repository
- AgentShield: unclear if self-built or third-party
- NanoClaw v2 lacks independent documentation or API reference
- ccg-workflow dependency: specific repo URL and version not provided in materials