When you're building AI agents that need to understand their own history across Claude Code, Codex, and Cursor, this gives them a standard way to query session files without hardcoding paths or JSON shapes for each platform. Point it at a repo name and a time window, optionally filter by keywords, and it emits normalized JSONL metadata. The keyword filtering is smart: it counts matches per session and adds relevance scores, which saves you from writing grep loops when you want to rank sessions by topic. It's infrastructure for meta-agents, the kind that reflect on past conversations to inform current work. If you're not building agents that examine their own session history, you won't need this.
npx -y skills add everyinc/compound-engineering-plugin --skill ce-session-inventory --agent claude-codeInstalls into .claude/skills of the current project.
Select a file.
juliusbrussee/caveman
mattpocock/skills
shadcn/improve
obra/superpowers
forrestchang/andrej-karpathy-skills
vercel-labs/skills