← 返回 Skills 市场
leostehlik

Proof Loop

作者 Leo Stehlik · GitHub ↗ · v0.1.0 · MIT-0
cross-platform ✓ 安全检测通过
52
总下载
0
收藏
0
当前安装
1
版本数
在 OpenClaw 中安装
/install proof-loop
功能描述
Multi-agent sprint protocol that prevents AI coding agents from declaring done without proof. Enforces spec freeze before build, role-separated subagents (bu...
使用说明 (SKILL.md)

Proof Loop

A sprint is not done until every acceptance criterion has a PASS verdict from a fresh verifier session.

Read references/workflow.md for the full loop spec. Read references/brief-template.md for the agent brief format. Read references/artifacts.md for the artifact schema.

The Loop

spec freeze -> build -> evidence -> FRESH verify -> fix -> FRESH verify
                                         ^                      |
                                         |______________________|
                                         (repeat until all ACs = PASS)

Four Roles — Always Separate

Role Does Never
Spec-Freezer Writes spec.md with explicit ACs Edits production code
Builder Implements against frozen spec Verifies own work
Verifier Fresh session — verdicts each AC Edits production code
Fixer Minimal fix for what verifier flagged Signs off on completion

The verifier is always a fresh session. The agent that built cannot judge its own work.

Acceptance Criteria Format

Every sprint brief must include explicit ACs before build starts:

AC1: [specific, testable condition — not a task description]
AC2: [specific, testable condition]
AC3: [specific, testable condition]

Good: "AC1: A German-locale user sees all prompt form field labels in German" Bad: "AC1: Translate the form fields"

Sprint is DONE Only When

  • Every AC has a PASS verdict in the verifier's verdict.json
  • No problems remain in problems.md
  • Full regression suite passes (if applicable)

Artifacts (stored in repo)

.agent/tasks/\x3CTASK_ID>/
  spec.md         -- frozen ACs + constraints + non-goals
  verdict.json    -- AC verdicts per phase (PASS/FAIL/UNKNOWN)
  problems.md     -- specific failures with file/line refs (if any)

See references/artifacts.md for schemas.

安全使用建议
Treat this as an incomplete low-confidence review: installation should wait until metadata.json and the artifact directory can be inspected directly.
能力标签
cryptocan-make-purchases
能力评估
Purpose & Capability
No supplied artifact evidence was available to show a purpose-capability mismatch.
Instruction Scope
No supplied artifact evidence was available to show unsafe or deceptive instruction scope.
Install Mechanism
No supplied artifact evidence was available to show a risky install mechanism.
Credentials
No supplied artifact evidence was available to show disproportionate environment access.
Persistence & Privilege
No supplied artifact evidence was available to show unsafe persistence or privilege behavior.
如何使用
  1. 确保已安装 OpenClaw(本地或 Docker 部署)
  2. 在对话框中输入安装命令:/install proof-loop
  3. 安装完成后,直接呼叫该 Skill 的名称或使用 /proof-loop 触发
  4. 根据 Skill 的参数说明提供必要输入,即可获得结构化输出
版本历史
v0.1.0
Initial public Proof Loop skill.
元数据
Slug proof-loop
版本 0.1.0
许可证 MIT-0
累计安装 0
当前安装数 0
历史版本数 1
常见问题

Proof Loop 是什么?

Multi-agent sprint protocol that prevents AI coding agents from declaring done without proof. Enforces spec freeze before build, role-separated subagents (bu... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 52 次。

如何安装 Proof Loop?

在 OpenClaw 或 Claude Code 对话框中运行命令「/install proof-loop」即可一键安装,无需额外配置。

Proof Loop 是免费的吗?

是的,Proof Loop 完全免费,采用 MIT-0 许可证,可自由下载、安装和使用。

Proof Loop 支持哪些平台?

Proof Loop 跨平台运行,可在任意部署了 OpenClaw / Claude Code 的环境中使用(cross-platform)。

谁开发了 Proof Loop?

由 Leo Stehlik(@leostehlik)开发并维护,当前版本 v0.1.0。

💬 留言讨论