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/edwinhu/workflows --skill dev-explore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。
基于 SOC 职业分类
正在显示 SKILL.md
| name | dev-explore |
| description | Internal read-only reconnaissance for dev planning. |
| user-invocable | false |
| disable-model-invocation | true |
Announce: "I'm using dev-explore to map the relevant code and test path."
!bun ${CLAUDE_SKILL_DIR}/../../scripts/load-constraints.ts dev-explore
The opening dev clarification sentinel must authorize this session before any discovery. Do not write planning artifacts. Do not design or implement.
Use targeted, parallel read-only exploration when it helps. Trace the actual entry point, data flow, integration boundaries, relevant conventions, similar implementations, and the test infrastructure. Read every key file selected from exploration; summaries are not a substitute.
Return these findings directly in the conversation:
path:line entries and their purpose.Iron law: a test must do what the user does. Testing an alternate HTTP path when production uses WebSocket, or inspecting source/logs instead of exercising runtime behavior, is not evidence. If no usable automated harness exists, identify setup as the first planned task and obtain the user's decision; never treat absence of tests as a waiver.
When reconnaissance exposes uncertainties, immediately load skills/dev-clarify/SKILL.md. Do
not write EXPLORATION.md, update a spec, or create a handoff/ledger.