Control plane for AI agents on a portfolio of repos: memory with provenance, scoped permissions, audit, a signed work queue, and a self-improvement loop scored against hidden tests. Cloudflare Workers + D1.
4 files for Claude Code: PreToolUse, Stop, capsid-mcp, capsid-mcp CLAUDE.md — 2,409 tokens loaded in every session.
CLAUDE.md A 2,409 tok .claude/settings.json C — .claude/settings.json A — .claude/settings.json A — These files are DrDustinEdwards/capsid-mcp's own configuration — they tell Claude Code how to work on this repository, so they are not mods to install elsewhere. Copy one as a starting point and replace the parts that are about this project.