OpenWolf

Appearance

Commands ​

Complete reference for the OpenWolf CLI.

openwolf init ​

Initialize OpenWolf in the current project.

bash
openwolf init                          # auto-detect installed agents (default)
openwolf init --agent codex opencode   # wire exactly these agents
openwolf init --agent all              # wire every supported agent
openwolf init --agent claude           # Claude Code only, skip detection

What it does:

  1. Detects the project root (.git, package.json, Cargo.toml, and so on)
  2. Creates .wolf/ with the state files, the durable index, and a .gitignore splitting committed state from machine-local runtime
  3. Copies the hook scripts to .wolf/hooks/ and registers 12 hooks in .claude/settings.json
  4. Auto-detects other installed agents (Codex, OpenCode, Gemini CLI, Cursor) and wires each one
  5. Installs the skills (/handoff, /security-audit, /reframe, /designqc) for every wired agent, plus the openwolf protocol skill for Claude Code
  6. Writes a five-line CLAUDE.md stub and .claude/rules/openwolf.md
  7. Runs the initial anatomy scan (descriptions, symbols, import graph)

Re-running init is safe: templates refresh, learned data is preserved, and existing hook entries in .claude/settings.json are merged, not replaced.

openwolf status ​

Health, stats, and file integrity: core files present, all hook scripts present, agent registrations, token stats, index size, daemon state.

openwolf scan ​

Force a full anatomy rescan.

bash
openwolf scan

Rescans descriptions, token estimates, tree-sitter symbols, and the import graph. Lockfiles, caches, minified files, and agent-config directories are excluded automatically. Normally you never need this: the post-write hook updates entries incrementally, and the daemon rescans when the index is actually stale.

openwolf scan --check ​

Exits 1 if the index no longer matches the tree. CI-friendly:

bash
openwolf scan --check || echo "index out of date"

openwolf find <query> ​

Locate a symbol or file from the index alone. Results are ranked by match quality, then import-graph importance, and the output is capped near 1,000 tokens so agents can use it instead of grepping the tree.

bash
openwolf find validateToken
src/auth/token.ts:82-140 method Auth.validateToken ~450 tok
src/auth/token.ts:5-160 class Auth ~1,240 tok
src/middleware/verify.ts file ~380 tok Token verification middleware

openwolf find --file <path> ​

Full index detail for one file: description, size, importance, and every symbol with its line range. The cheap replacement for reading the file, or the index, whole.

openwolf map ​

A token-budgeted overview of the most important files, ranked by personalized PageRank over the import graph. The ranking is seeded by files your recent sessions touched and by --focus terms, then fitted to the budget by binary search.

bash
openwolf map                    # ~1k tokens (2k when no session seeds exist)
openwolf map --focus auth,jwt   # bias the ranking toward these terms
openwolf map --budget 500       # explicit output budget

openwolf report ​

The token report, hardest numbers first:

  • Measured: real usage scanned from every project transcript right now, per model, subagent sidechains included
  • Bash governor: original output versus what entered context, measured at the rewrite point
  • Cache rebuilds: the last 7 days of prompt-cache invalidations with their triggers (model switch, compaction, version change, expiry) and token cost
  • Estimates: clearly labeled heuristics, including OpenWolf's own injection cost
bash
openwolf report

openwolf bench ​

The A/B benchmark: the same task set against two fresh clones of a fixture repo, one with OpenWolf and one bare, via headless runs.

bash
openwolf bench --repo /path/to/fixture --yes
openwolf bench --repo <git-url> --task bugfix --repeats 5 --yes

Reports medians per token dimension (input, output, cache read, cache write), task completion, and the bash re-run rate. Spends real API budget; refuses to run without --yes. Raw results are written to a JSON file.

openwolf bug search <term> ​

Full-text search over the bug memory, relevance ranked (SQLite FTS on Node 22.5+, substring fallback below).

bash
openwolf bug search "cannot read properties"

openwolf dashboard ​

Open the dashboard. Starts the daemon automatically if it is not running; no PM2 required. Each project gets its own port and token.

openwolf daemon ​

bash
openwolf daemon start     # persistent daemon via PM2
openwolf daemon stop      # stops PM2 or forked daemons alike
openwolf daemon restart
openwolf daemon logs      # last 50 lines

The daemon handles stale-gated anatomy rescans, measured-usage refresh, memory consolidation, and the dashboard server. It makes no network calls.

openwolf cron ​

bash
openwolf cron list        # tasks, schedules, last runs
openwolf cron run <id>    # trigger now (works without the daemon)
openwolf cron retry <id>  # clear a task from the dead letter queue

openwolf update ​

Update every registered project to the installed OpenWolf version.

bash
openwolf update                   # all projects
openwolf update --project my-app  # one project (partial match)
openwolf update --dry-run         # preview
openwolf update --list            # show registered projects

Each update takes a timestamped backup first, merges new config defaults without touching your values, removes dead weight only when it is verifiably untouched template content, and then verifies the hook install with a per-hook selfcheck. A failed verification fails the update loudly instead of leaving a broken install.

openwolf restore [backup] ​

List backups (no argument) or restore .wolf/ from one:

bash
openwolf restore
openwolf restore 2026-08-20T1655

openwolf --version ​

bash
openwolf --version

AGPL-3.0 · Made by Cytostack

Copyright 2026 Cytostack Pvt Ltd