OpenWolf

OpenWolf is open-source middleware for Claude Code (github.com/cytostack/openwolf). It installs six Node.js hook scripts that intercept Claude's file read/write lifecycle, maintaining a persistent project index (anatomy.md) and learning memory (cerebrum.md) in a .wolf/ directory. Result: 65.8% average token reduction across 20 projects by preventing redundant file reads and short-circuiting reads with cached summaries.

Reference Files

Load on demand based on the question:

• setup-and-architecture.md — Install steps, Node.js requirements, what openwolf init does, hook registration in .claude/settings.json, overall data-flow diagram, token estimation methodology. Load for setup, "how does it work", or architecture questions.
• wolf-directory.md — Every file in .wolf/: purpose, format, when updated, and the markdown-vs-JSON state principle. Load when a user asks about a specific file, wants to edit .wolf/ manually, or is debugging stale state.
• hooks-reference.md — All six hooks: event type, script name, behavior, inputs/outputs, timeout, and failure mode. Load for hook debugging, understanding what fires when, or customizing hook logic.
• cli-and-config.md — Full CLI command reference (init, status, scan, dashboard, daemon, cron, designqc, bug, update, restore), config.json schema, exclude patterns, cron schedules, dashboard port. Load for CLI questions or config tuning.
• gotchas.md — Licensing (AGPL-3.0), Node.js-only hooks, cerebrum compliance limits, token estimation accuracy, stale anatomy, large directory exclusions, Claude Code exclusivity, daemon/PM2 setup. Load when troubleshooting or evaluating adoption risk.

Quick-Start Decision Flow