rig-lite is a portable governance kit for AI coding agents, shipped by the Turbo Flow project. It is a set of bash and Python files you drop into any repository: gate.sh, a fail-closed cross-model review gate; constitution.md, the rules agents operate under; a git-versioned memory pattern; worktree isolation for parallel writers; and a self-test suite that proves the kit fails closed. It ships no harness and runs no daemon.

Does rig-lite work with Claude Code, Codex, or other agents?

Yes. The kit is harness-agnostic: any tool that reads AGENTS.md and can run bash works as a builder — ZCode (GLM), Claude Code, Codex, Gemini, or Grok. The gate knows the builder families and refuses same-family review, so the reviewer must come from a different model family. Orchestration like Ruflo is optional and installed separately.

How long does it take to set up rig-lite?

About ten minutes with no new dependencies: run init-repo.sh from the repo you want to govern (it writes a thin AGENTS.md and never clobbers existing files), run self-test.sh to prove the kit on your machine, then gate your first pull request with gate.sh --pr 1 --builder . Everything is bash.

~/turbo-flow/rig-lite MIT · bash · zero deps

Add a cross-model code-review gate to Claude Code — or any coding agent

rig-lite is the portable governance kit extracted from Turbo Flow. Your agents already plan, code, test, and revise — the kit adds the layer that decides whether any of it is allowed to ship: a fail-closed cross-model review gate, a written constitution, git-versioned memory, and a human holding the merge button.

Published Oct 7, 2026 · Updated Oct 8, 2026

no daemon no orchestration platform no vendor lock-in bash + python, that's it works in any repo

The parts

file	what it gives you
gate.sh	the cross-model review gate — deterministic checks first (they're free), then a reviewer from a different model family must return a parseable `APPROVED` / `REVISE`. Fail-closed; never merges. `--pr <n>` gates a GitHub PR and posts the verdict as a comment; `--sweep` gates every open PR without a verdict yet; `-C <repo>` gates another checkout. Every verdict lands in a local JSONL log.
constitution.md	the laws — builder ≠ reviewer, agents never merge, worktrees for parallel writers, secrets encrypted, state on disk. Ships as a fill-in template.
memory.md + memory/	the git-versioned memory pattern: index, decisions, gotchas, project index, inbox. Agents forget; git doesn't.
self-test.sh	the kit's own fail-closed test suite — hundreds of checks. It proves the gate fails closed before you trust it with a real PR.
wt.sh	an isolated worktree per parallel writer — `wt <name>` creates the worktree + branch, `--clean` tidies up after merge.
init-repo.sh	one-command onboarding for any repo: a thin `AGENTS.md` (constitution pointer + cheat-sheet) and the `CLAUDE.md → AGENTS.md` symlink. Never clobbers existing files.
specs/	spec + UAT contract templates — plan before build, verify behavior against the deployed app.
digest.sh	fleet health in one file: merge queue joined with gate verdicts, dirty checkouts, site liveness, stale codespaces.
automations/	nine scheduled tasks as a portable prompt pack — morning digest, security scan, memory consolidation, weekly retro, and more.
secret.sh	encrypted-at-rest secrets with backend auto-detection (Keychain → libsecret → age → loudly-warned chmod-600). Listing shows names, never values.
tokens.py	a live token-burn dashboard across zcode + claude + codex + gate. Read-only by construction — the first open-core drop from Turbo Rig.
hooks/pre-commit	deletion guard: staged deletions print loudly instead of silently vanishing. Warn-only, fail-open.

Bring your own harness

The kit ships no agent and no orchestrator. It names harnesses by role and works with anything that reads AGENTS.md and can run bash:

role	who	how the kit uses it
Builder	ZCode (GLM), Claude Code, Codex, Gemini, Grok	`gate.sh --builder <cli>` knows the family; anyone can build
Reviewer	Claude Code or Codex CLI — never the builder's family	the gate auto-detects and refuses same-family review
Orchestration	optional — Ruflo as a Claude Code mod	bring it if you want swarms; the kit doesn't care

Ten-minute install

# from any repo you want to govern:
bash /path/to/turbo-flow/rig-lite/init-repo.sh   # writes AGENTS.md + symlink, never clobbers
bash /path/to/turbo-flow/rig-lite/self-test.sh   # prove the kit on your machine
/path/to/turbo-flow/rig-lite/gate.sh --pr 1 --builder zcode

# no Codespace? fine — it's bash. want the container tour anyway?
# open the repo in a Codespace; postCreate runs the self-test.

Why trust it

The gate survived 17 rounds of review by a model family that didn't write it — REVISE'd seventeen times before earning APPROVED, counted from the gate's own JSONL log. The findings were exactly the class the gate exists to catch: a prompt-injection hole that let a diff approve itself, a fail-open path that skipped review, a quoting bug that swallowed failing tests.
Every fix became a self-test. 252 checks at v5.2, all passing under bash 5.3 and stock macOS 3.2; shellcheck clean at warning severity.
Fail-closed is the design, not a setting. Ambiguity is REVISE; the gate never merges; the merge button is a human's. See the philosophy.

Where the rest lives

rig-lite is the portable extract of Turbo Rig — the same gate, constitution, memory and automations running as one integrated, always-on system. Turbo Rig is in private beta. The story of how the kit came to be is here.