everything-claude-code

Overview

everything-claude-code is the harness-optimization bundle that became the most-watched agent-stack repo of Q2 2026. Built at the Claude Code Hackathon (Cerebral Valley × Anthropic, February 2026) by Affaan M, it consolidates four operator concerns — skills, instincts, memory, and security — into a single repository that drops onto any Claude Code installation and, increasingly, any compatible harness (Codex, OpenCode, Cursor).

The repo crossed 140,000 stars in early May 2026 and is gaining at roughly +1,000/day. That puts it inside the top ten harness-and-skill repos by star count, and inside the top five by velocity. The breadth of contribution — 170+ contributors across 12+ language ecosystems — is the second derivative tell that the project has crossed from “founder’s repo” to “community standard.”

Key Capabilities

The bundle is organized around four overlapping layers:

• Skills: A curated skill pack engineered to plug into Anthropic’s official Skills surface, but also exported in formats consumable by Codex, OpenCode, and Cursor. The skills are explicitly opinionated about quality gates rather than just task coverage.
• Instincts: Persistent behavior-modification layer. Where a single CLAUDE.md captures rules, the instincts module captures the result of repeated corrections — the agent learns not just what to do, but what not to do based on prior session traces.
• Memory: Cross-session persistence layer that captures what the agent learned and surfaces it back at session start. Sits alongside (rather than replacing) Anthropic’s MEMORY.md convention.
• Security: A dedicated configuration scanner that checks Claude Code installs (and adjacent harnesses) for vulnerabilities and misconfigurations. 102 static-analysis rules cover secrets detection (14 patterns), permission auditing, hook injection analysis, MCP server risk profiling, and full agent-config review.

Why It Matters

The security module is the part most operators underestimate. As of mid-2026, hostile-skill installation is a real attack surface for agentic dev environments, and the validator wave (react-doctor, claude-doctor, agentmemory’s benchmarks) has been mostly about output quality. everything-claude-code is the first widely-starred project that treats the configuration itself as a thing to validate — checking for leaked secrets in CLAUDE.md, overscoped permissions, and risky MCP server connections before they become an incident.

Best Fit For

• Operators running multi-harness setups (Claude Code + Codex + OpenCode) who want a single optimization layer that travels across
• Teams that need a security audit step before shipping skill packs to production
• Power users who already invested in CLAUDE.md / skills / memory and want to consolidate

Installation Notes

The project is MIT-licensed and runs as a drop-in addition to existing Claude Code installations. The README covers the four-layer setup in order; for first installs, the security scanner alone is worth the install path even if you don’t adopt the skills or instincts layers.

Related Reading

• Skill Spam Is a Genre — And the Validators Are Trending
• mattpocock vs Composio — the skills directory race