用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt;运行前请先检查来源。
npx skills add https://github.com/mugsun/curdx-flow --skill prompt-optimize命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地?可下载 SkillsMP 当前能够提供的文件。
正在显示 SKILL.md
| name | prompt-optimize |
| description | Use when optimizing a task prompt or asking how to route curdx-flow work without executing it. |
| argument-hint | [draft prompt or task goal] |
| allowed-tools | Read Bash |
| disable-model-invocation | true |
Advisory only. Optimize the user's task prompt and recommend curdx-flow routing. Do not execute the task, edit files, create specs, run implementation commands, spawn executor agents, commit, push, or tag.
$ARGUMENTS: draft prompt, task goal, or rough idea to improve.Run the deterministic router in read-only mode:
curdx-flow route --goal "$ARGUMENTS"
Use the returned route, intent, stackProfile, qualityGates,
suggestedVerifier, contextBudget, and recommendedCapabilities as facts.
For non-trivial routes, also run:
curdx-flow route --compile --goal "$ARGUMENTS"
Use the execution brief to describe context budget, completion evidence,
agent/reviewer plan, and .curdx/brain.jsonl warnings. Do not execute the
brief.
If the command is unavailable, infer conservatively from the prompt and state
that routing is estimated.
Respond in the user's language with exactly these sections:
## Prompt Diagnosis
- Strengths:
- Gaps:
## Recommended Route
- route:
- why:
- context budget:
## Recommended curdx-flow Skills/Agents
| Item | Use |
|---|---|
## Risks
- ...
## Quality Gates
| Gate | Required | Command |
|---|---:|---|
## Optimized Prompt
```text
...
```
stackProfile.primary != "unknown".direct-change: prompt should ask for a small direct edit plus exact verifier.lite-spec: prompt should ask for a lightweight spec and 1-3 value-slice tasks.full-spec: prompt should ask for research -> requirements -> design -> tasks
-> implementation with phase verification.epic-split: prompt should ask for /curdx-flow:triage.scaffold: prompt should require latest official/ecosystem scaffold docs and
baseline verification.product-inception: prompt should ask for product context before source code.greenfield-spec: prompt should ask for product context, constitution, walking
skeleton, then vertical slices.prototype: prompt should include a bounded success criterion.import-spec: prompt should preserve traceability to the source artifact.