跳到正文
cccskills
免费在 GitHub 公开

crack-plan

Start a substantial build with Codex on Crack. Inspect the user's own configured models and readiness, recommend a lead/worker/tool shape with reasons, and let the user decide before work starts. Use before delegating; keep small or solo tasks direct.

中文简介正在准备中。当前显示原文说明。

查看安装方法

包含的文件(3)

  • SKILL.md3.2 KB
  • agents/openai.yaml352 B
  • scripts/plan.mjs11.9 KB

SKILL.md(原文)

安装前可以查看交给 AI 代理的指令内容。

Plan one build, then choose

Use this when the user wants to start a meaningful build and has not already decided how to divide it. The goal is a short, honest plan built from what is actually configured, with the final choice left with the user.

Resolve named models before recommending

A request such as “run subagents through Opus 5.5” already selects a model. Follow the model-request routing in crack; the user need not specify the subscription or adapter. The native planner below does not inventory Claude Code access, so a missing native role is not proof that Opus is unavailable. Check the external adapter separately when requested. Keep route ambiguity, missing access, and unsupported model versions explicit.

Gather local facts only

Run node scripts/plan.mjs (add --task implementation|ui|review|research|tests when the work is known). It reads the model catalog Codex already uses, the generated roles, and static readiness. It makes no model call, changes no configuration, and never probes a provider.

Recommend with reasons, then let the user decide

When several configured workers qualify, offer the recommendation the planner returns and say why in its terms: a role whose configured tasks list includes this category, a conventional role name for the task, and a generated file that is present and unmodified. Label that as a qualitative judgement about fit, not a measured ranking, and accept a different choice without argument.

If the user already named a worker role or a mode, honor it: do not re-ask for a decision that is already made. If they named something that does not qualify, say exactly which requirement failed rather than substituting another model.

Keep the roles distinct

  • The Codex host model is whatever the user selected in the composer. A skill cannot change it.
  • A configured worker implements one bounded bundle and is reviewed.
  • The optional external project lead is a different thing: the official Claude Code client on an existing subscription login, on the exact model claude-opus-5-5, with no Codex tools and no API fallback. Read ../crack/references/external-lead.md before offering it.

Host-only solo work needs no worker. Explicit Opus implementation uses the external adapter's solo mode and its readiness checks. When no configured worker qualifies, say so and suggest $codex-on-crack:crack-setup; do not invent a substitute.

Hand off

Once the shape is agreed, hand the work to $codex-on-crack:crack with the chosen route and worker, or keep a host-only task direct. For replay of a host session and worker sessions, read ../crack/references/replay.md.

Keep the claims honest

  • Catalog presence is not account access, provider routing, or quality.
  • No allowance, quota, savings, or price claims. Missing usage stays unknown.
  • Terminal tooling is an allowlist, not filesystem isolation; scope follows the approved workspace and task.

评价

还没有评价。欢迎分享你的使用感受。

评论

无需登录即可发表。欢迎分享使用感受、提问,或说明在哪些环境下无法运行。

还没有评论。来写第一条吧。

提交后会立即公开。请勿填写邮箱、电话等个人信息。URL 最多 2 个。

0 / 1000 字

此表单受 reCAPTCHA 保护,适用 Google 的隐私权政策和服务条款。

同一仓库的技能

Plan and execute substantial multi-file builds, features, migrations, and refactors with Astra as orchestrator and DeepSeek V4.1 Flash as the native Codex implementation worker. Use for phased planning and delegated build execution, including existing Superpowers or GSD plans. Skip trivial edits and explicitly single-agent tasks. Worker children must not invoke this orchestration skill.

中文简介正在准备中。当前显示原文说明。

ethanplusai/codex-on-crack7612026年10月4日 更新

crack

免费

Lead substantial Codex builds with one configured implementation worker and lead acceptance. Use when the user requests codex-on-crack, names a model to delegate work to, asks to run subagents through Claude, or requests a lead/worker split. Keep host-only work direct; honor explicit external model choices. Workers must not invoke this orchestrator.

中文简介正在准备中。当前显示原文说明。

ethanplusai/codex-on-crack7612026年10月4日 更新

Work with the opt-in codex-on-crack build panel. Report registered run and review status, submit new screenshot evidence for a registered review, and send the user to the panel to decide. Never record user feedback or visual approval yourself.

中文简介正在准备中。当前显示原文说明。

ethanplusai/codex-on-crack7612026年10月4日 更新

Configure or change the approved worker models for codex-on-crack, inspect available models and compatibility, preview role files, or undo a setup. The user selects the lead in Codex. Setup preserves root settings, credentials, unrelated agents, and config.toml.

中文简介正在准备中。当前显示原文说明。

ethanplusai/codex-on-crack7612026年10月4日 更新

查看 ethanplusai 的全部技能

报告此技能的问题