A self-maintaining 3-level memory system for Claude Code. Give your AI an org chart and a memory that compounds. Stdlib only, no API key.
Command
Model-driven lesson capture — read recent work in a role, distill genuine insights, append tagged bullets to pending.md. The real capture path; complements the Stop-hook's cheap regex fast-lane.
Command
Scaffold a new project under an existing role — creates the folder, memory/PROJECT.md, updates roles.json, and regenerates ORG.md.
Command
Render a shareable org-map SVG (ORGMAP.svg) of this repo's org — roles, projects, captured lessons, token savings. Built to screenshot and post.
Command
Show the org at a glance — roles, folder counts, drift (unmapped/missing folders), adoption progress, and which role brains are stale or have pending activity due for /role-promote.
Command
Intelligent, read-only preview of how orgkit would organize THIS repo. The model reads your actual folders (not keywords), proposes a sensible org with rationale and an estimated token saving, and writes nothing. The "look before you leap" twin of /orgkit-init.
Command
Analyze your real Claude Code usage and recommend a capture cadence + time slot, then optionally install a low-impact background batch-capture cron. Read-only analysis; the cron is opt-in.
Command
Diagnose and interactively fix broken orgkit state — verifies the .org/ engine, roles.json, role brains, hook registration, and ORG.md, then offers to repair what is broken. Pass --dry-run to report only without fixing.
Command
Intelligently onboard the current repo into orgkit. The model reads real folder signals, reasons over content (not keywords), proposes a sensible org structure, lets the user refine it, then writes .org/roles.json and scaffolds the memory system.
Command
Safely migrate existing folders into role directories — dry-run preview first, auto-fix literal path strings, then the model handles import statements and relative paths it flags. Never claims success silently.
Command
Autonomously reconcile a role's memory — merge pending + recent learnings into ROLE.md, dedupe and declutter. The auto-promotion step fired when a role is stale (>7 days) with pending activity.