chouswei
chouswei/cursor-user-skills
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/chouswei/cursor-user-skills --skill engineering-practices-learner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。
Keep a working plan as MemNet session rows (campaign or plan task plus ordered step tasks with execution waves). Pin the live neighbourhood, then mutate to draft, update, or repolish. At plan time, mark which steps may run in parallel. Execute a ready wave via memnet-multitask when asked. Chat is a Shape of the graph, never the plan SSOT. Triggers: memnet plan, plan in memnet, update memnet plan, repolish plan, session plan, memnet planner, plan waves, parallel plan steps, execute plan wave. Skip: Markdown-only project-planner interview with no MemNet; building the MemNet engine.
Enforceable MemNet doctrine for Cursor Multitask Mode and Task sub-agents: one shared session SSOT, TCP or streamable-http transport, parent/worker MUST/MUSTNOT, MN-REQ-12 usage, system-dev two-store pattern for modelbasedPrj-*. Triggers: Multitask Mode, multitask, multi-agent, Task sub-agent, background worker, parent coordinator, delegate worker, shared session, memnet multitask, system-dev multitask, modelbasedPrj multitask, MN-REQ-12, parallel workers, TSK_* settle, TCP serve, streamable-http MCP, GQL wire, shaped pin_map.
How to use MemNet as mission working memory: goldfish loop, chat never SSOT, drop prior maps. Triggers: use memnet, how to use memnet, memnet goldfish, mission working memory, chat never SSOT, session graph.
基于 SOC 职业分类
正在显示 SKILL.md
| name | engineering-practices-learner |
| description | >- |
Role: Structure practices with labels/tags/taxonomy and typed relations between entries (not a flat tag soup only).
Run pipeline_steps; do not skip step 3 when retrieval matters.
Resources: references/core-engineering-practices.md · assets/engineering-practices-output-template.md
Step 3 tool: python tools/engineering-practices-retriever.py "<query>" or ADK engineering-practices-retriever.
Pairing: scientific-method-first-principles when step 4 sets method_check: true; skill-reviewer then skill-creator when promoting practices to a repo skill (see below).
A coherent subset of classified practices (stable ids, clear conditions, solid observed results) can become a new Agent Skill. Typical fit:
tool-wrapper when the bundle is conventions + when to apply (load references/ and follow when coding or reviewing); optional small tools/*-retriever.py over a trimmed principles file copied from this output.pipeline only if the workflow stays ordered steps (gates, JSON phases); not required for most practice libraries.Review before scaffold: Treat the filled step-5 template (tables + relations) as the source spec. Run skill-reviewer on the draft folder after skill-creator emits it, or ask skill-reviewer to audit the template export as a virtual skill spec (ids, safety, triggers). Then skill-creator ingests that spec to generate SKILL.md + references/ + assets.
Does not auto-write disk: This skill only structures knowledge; skill-creator creates files.