Godspeed

Godspeed

**A Claude Code workflow trigger backed by Toke — a routing classifier and multi-agent orchestration engine.**

[](https://opensource.org/licenses/MIT) [](https://docs.claude.com/en/docs/claude-code)

Type `godspeed` in a Claude Code session, the Toke engine activates for the rest of the turn: each prompt is scored on complexity (S0–S5), routed to the cheapest model that will handle it well, and — for complex tasks — decomposed into parallel Sonnet workers whose synthesis is critic-gated before it lands in a self-improving memory.

Zero fork of Claude Code. Everything wires in through the hook API.

**Free by default — no API key required.** Godspeed never auto-fires Anthropic API calls. The 5 lifecycle hooks use only the local-only `brain_hook_fast.js` classifier (a regex/keyword scorer running in your shell, not a network call). The Homer pantheon dispatches subagents through Claude Code's in-session `Agent` tool, which is part of your existing Claude Code session — no separate billing channel against your Anthropic key. Two paths can opt into direct API calls, both gated behind explicit user commands and an `ANTHROPIC_API_KEY` env var: `brain advise ""` (advisor escalation) and `agent_runner.py invoke --mode live` (live subagent dispatch). Unset the key, never type those commands, and the plugin runs at $0 against your Anthropic API account. The included `cost_guard.py` enforces per-tier USD ceilings on the live path so even when you opt in, you can't accidentally burn your credit balance.

---

What you get

Cost	Sonnet auto-routing on subagent calls via CLAUDE_CODE_SUBAGENT_MODEL=sonnet, plus per-tier USD ceilings (automations/homer/cost_guard.py) on the opt-in live-dispatch path. Real savings = your-Opus-spend × your-S3+-rate. Run brain scan after a few