이 기술은 OpenClaw-RL: Train Any Agent Simply by Talking에서 개발되었으며, 검증된 접근법을 레포-로컬 워크플로우에서 에이전트 플로우에 적용합니다.

"차세대 국가 신호는 보편적이며, 정책은 동시에 모든 신호로부터 배울 수 있다"고 말합니다.

레포 작업 증명 루프는 사소하지 않은 코딩 작업을 위한 레포-로컬 워크플로우 스킬입니다.

.agent/tasks/ / 아래에 내구성 있는 작업 폴더를 생성하고, 프로젝트 범위별 Codex 및 Claude 서브에이전트를 설치하며, 저장소 지침을 업데이트하고, 엄격한 루프를 실행합니다:

spec freeze -> build -> evidence -> fresh verify -> minimal fix -> fresh verify

Codex의 경우, 이 루프는 작업이 깔끔하게 분할될 때 explorer 또는 worker 자식에 대한 적응형 유명 팬아웃 경로도 지원합니다.

요점은 간단합니다: 증명을 저장소 내에 보관하고, 구현과 검증을 분리하며, 작업 상태를 나중에 쉽게 재개하거나 감사할 수 있도록 하는 것입니다.

그것이 만드는 것들

목표 저장소 내부:

또한 관리된 워크플로우 블록을 다음과 같은 곳에 삽입합니다:

• 레포 루트 AGENTS.md 코덱스 기준선
• 저장소의 Claude 가이드 파일: CLAUDE.md 또는 .claude/CLAUDE.md

설치

스킬을 프로젝트 스킬로 설치하세요.

코덱스

클로드 코드

Repo Task Proof Loop

This skill was built from OpenClaw-RL: Train Any Agent Simply by Talking and applies its proven approach to agentic flows in a repo-local workflow.

“next-state signals are universal, and policy can learn from all of them simultaneously.”

Repo Task Proof Loop is a repo-local workflow skill for non-trivial coding tasks.

It creates a durable task folder under .agent/tasks//, installs project-scoped Codex and Claude subagents, updates repo guidance, and drives a strict loop:

spec freeze -> build -> evidence -> fresh verify -> minimal fix -> fresh verify

For Codex, that loop also supports an adaptive bounded fan-out path for explorer or worker children when the task splits cleanly.

The point is simple: keep proof inside the repository, separate implementation from verification, and make task state easy to resume or audit later.

What It Creates

Inside the target repository:

.agent/tasks//
  spec.md
  evidence.md
  evidence.json
  raw/
    build.txt
    test-unit.txt
    test-integration.txt
    lint.txt
    screenshot-1.png
  verdict.json
  problems.md

.codex/agents/
  task-spec-freezer.toml
  task-builder.toml
  task-verifier.toml
  task-fixer.toml

.claude/agents/
  task-spec-freezer.md
  task-builder.md
  task-verifier.md
  task-fixer.md

It also inserts managed workflow blocks into:

• the repo-root AGENTS.md Codex baseline
• the repo’s Claude guide file: CLAUDE.md or .claude/CLAUDE.md

Install

Install the skill as a project skill.

Codex

.agents/skills/repo-task-proof-loop/

Claude Code

.claude/skills/repo-task-proof-loop/

If you use both tools on the same repository, install it in both locations or keep one canonical copy and sync it.

Quick Prompts

Use this prompt for the normal flow:

Do Task

Use $repo-task-proof-loop to do the task described below in this repository. Reuse the matching repo-local task if it already exists; if not, initialize it first and then continue automatically after init completes. You are explicitly authorized to use subagents and bounded parallel helper work when it materially helps. Choose the best internal orchestration automatically from the current task shape and tool surface. Keep the proof-loop phase explicit as you work so matching project agents can be picked automatically when the product supports that, otherwise continue on the main thread. Keep the task tree shallow, keep one integration builder responsible for evidence, and keep every verifier pass fresh.
...

For all prompts, replace ... with either Task file: on the next line or the task text pasted on following lines.

This skill is intentionally proof-first, so init always comes before build.