Ccdawn goal loop

Skill CCDawn/codex-skills/skills/engineering/ccdawn-goal-loop

Chinese-first Codex skill pack for intent alignment, dynamic routing, lightweight development, code review, UI design, and AI research workflows.

Install

npx -y skills add CCDawn/codex-skills --skill ccdawn-goal-loop

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

4 stars4 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Use when the user explicitly requests an open-ended, persistent iteration policy with evidence-based next-step selection and stop conditions, and no domain-specific research, score, development, or coordination owner already controls the loop.

The file declares its own license as MIT. That is the author’s claim about this one file, and it is not the same thing as the license GitHub reports for the repository, which is listed with the other numbers below.

SKILL.md

2.9 KB, as published. Nobody here has run it

CCDawn Goal Loop

目标

为明确要求持续迭代的开放目标提供轻量控制策略。它不替代 BRT 对齐、专项 owner、Planning 或 Project Memory，也不用于普通有限任务。

BRT interface

Context Boundary: 已对齐目标、不可破坏约束、可用范围、当前证据和停止条件。
Output Contract: 当前最有价值动作、最新证据、ADVANCE / ITERATE / STOP / BLOCKED 和下一动作。
Allowed Action: 在用户授权和约束内执行最小高信息动作；不因本 skill 自动扩大工具、写入或远程权限。
Success Evidence: 当前动作产生能改变完成判断或下一选择的可检查证据。
Stop Condition: 目标达成、约束将被突破、没有高价值动作、连续尝试无新证据、缺输入/权限或用户暂停。
Route Out: 最具体专项 owner、继续当前 loop、ccdawn-brt 或 BLOCKED。

统一调用契约

只处理 BRT interface 范围；不匹配时回 ccdawn-brt 或更具体 owner，复合任务不吞其他 owner。
用户可见内容默认中文，完成只报状态、产出、证据和剩余风险；代码、命令、路径、错误原文、API/协议、skill 名和枚举保留原样；Route Out 仅以 BRT interface 为准，末行写 下一步建议: <一个具体动作>。

激活闸门

必须同时满足：用户明确要求持续/反复迭代；终点不能预先拆成有限任务；每轮需要根据证据选择下一动作；没有更具体 owner。普通“继续”“优化到完成”、有限实现、已知 task graph、实验/score/research loop 和多 Agent 协调均不触发。

最小循环

首轮只确认 Goal / Evidence / Constraints / Stop；BRT 已提供的内容直接继承，不重复询问或展示合同。
选择最小且最可能产生新证据的动作，执行并验证。
根据证据判断：推进目标用 ADVANCE，修正方法用 ITERATE，目标已达成或继续无价值用 STOP，缺不可约输入/权限用 BLOCKED。
已授权且未触发停止条件时连续迭代，不在每轮询问是否继续。

Goal Loop 是控制策略，不是状态存储。对话内续接使用 BRT Runtime；只有用户/项目已启用跨会话 memory 且当前决定值得持久化时，才由 ccdawn-dawn-agent-html-memory 写一次 durable delta。不得逐轮生成 ledger、dashboard、完整尝试历史或重复 goal contract。

输出

普通轮次只给：本轮结果 / 新证据 / 决策 / 下一动作。只有 BLOCKED 才补已尝试的关键路径和一个不可约问题；只有用户明确要求正式交接时才路由 Completion Summary。

下一步建议: <一个具体动作>