Lupus — Development skill for Claude Code
Local supervisor that makes Claude Code and Codex CLI finish verifiable work: evidence-only completion, frozen tests, budgets, crash recovery and Claude<->Codex handoff.
How to install Lupus
This entry records only its repository, not the path inside it, so there is no
exact command to give. Open djfksjd/lupus and copy the folder into
~/.claude/skills/, or the file into ~/.claude/agents/.
What Lupus does
Local supervisor that makes Claude Code and Codex CLI finish verifiable work: evidence-only completion, frozen tests, budgets, crash recovery and Claude<->Codex handoff.
Alternatives in Development
- Create Handoff — You are tasked with writing a handoff document to hand off your work to another agent in a new session 10k ★
- Lacp — Local-first policy, evidence, memory, and recovery controls for Claude, Codex, Hermes, and other CLI coding ag 300 ★
- Work Context Handoff — Generate a comprehensive handoff document capturing the current state of work for seamless continuation by ano 77 ★
README
LUPUS
**AI WORKS TOGETHER**
A local supervisor that makes Claude Code and Codex CLI finish verifiable work —
with evidence, budgets, crash recovery and Claude ↔ Codex handoff
[English](./README.md) · [한국어](./README.ko.md) · [简体中文](./README.zh-CN.md) · [日本語](./README.ja.md)
[](./LICENSE)    
Lupus runs the `claude` and `codex` CLIs you already have (with your existing subscription login) as workers, and decides "done" itself: a goal is complete only when checks the worker cannot influence pass, never because a model said so.
**This is `v0.2 alpha`.** A single-user tool for macOS. Workers and verifiers run in OS sandboxes, not in a VM. The measurements below are small. Do not use it on sensitive material.
What it does
| Evidence-only completion | DONE requires passing checks bound to the current acceptance criteria. A worker's "I fixed it" is not evidence. |
| Frozen tests | Test files and runner config are frozen before a worker starts. Edited, deleted or skipped tests are put back before verification. |
| Python, Node, Go, Rust — or any command | unittest, pytest, node:test, jest, vitest, go test, cargo test are recognised from the project's files. Anything else: --check "". |
| One line for any request | lupus fix-tests turns the red tests it observes into the goal. lupus do "" drafts a failing test (and, kept aside, a proposed implementation) |
Related Skills
Agent Looper
Fix-until-green loop for coding agents: frozen goal, shell verify as the finish line, fresh worker each round,
Verification Before Completion
Use when about to claim work is complete, fixed, or passing, before committing or creating PRs - requires runn
Explain Consumer Pick Divergence
Confirm whether the rota gateway and the handoff planner sending work to different silos is expected or a bug,
Recovery Toolkit
A native token counter that reads your local session logs — no network, no npx, no Node — plus three Claude Co
Autowonder Claude Executor
Open-source executor for AutoWonder community edition — the complete scheduler protocol surface (dispatch, rel
Metaskill Evolve
Open $0 reproduction + honest dissection of MetaSkill-Evolve (two-timescale recursive skill evolution). Frozen
Related Agents
Codex Debugger
Root-cause a failing test, crash, stack trace, or misbehaving feature by delegating the investigation to OpenA
Dual Reviewer
Second-phase Codex code-review loop with Claude adjudication, after pr-reviewer in the branch review pass or m
Review Crash Safety
Reviews code changes for crash safety, WAL correctness, durability guarantees, atomic operations, page-level c