菜单
Use when writing prompts, agent instructions, SKILL.md, commands, system prompts, Task tool prompts, prompt engineering, or LLM-to-LLM content
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。
基于 SOC 职业分类
Use when facing hard architectural decisions, multiple valid approaches exist, need diverse perspectives before committing, or want M-of-N synthesis on complex problems
Use when rough ideas need design before code, requirements are fuzzy, multiple approaches exist, or you need to explore options before implementation
Use when testing MCP servers, debugging MCP tool responses, exploring MCP capabilities, or diagnosing why an MCP tool returns unexpected data
Use when debugging bugs, test failures, unexpected behavior, or needing to find root cause before fixing
Use when managing project learnings - view, add, search, prune, or export operational knowledge that persists across sessions
Use when shipping code - complete workflow: merge main, run tests, multi-review with auto-fix, commit, push, create PR. Replaces manual commit-push-pr flows.
| name | writing-for-llms |
| description | Use when writing prompts, agent instructions, SKILL.md, commands, system prompts, Task tool prompts, prompt engineering, or LLM-to-LLM content |
| version | 1.0.0 |
| category | meta |
| triggers | ["prompt","write prompt","system prompt","agent instructions","SKILL.md","prompt engineering","Task tool"] |
Apply the full prompt engineering standards from @rules/prompt-engineering.mdc
- Show correct patterns only - never show anti-patterns, even labeled "wrong" - State goals, not process - trust the executing model's capabilities - Use XML tags for structure in complex prompts - Clarity over brevity - every word that adds clarity is worth including - Few-shot examples must follow identical structure - Front-load critical information - Consistent terminology throughout Before finalizing: - No anti-patterns shown anywhere - All examples structurally consistent - Goals clear, process not over-prescribed - Terminology consistent - Critical info front-loaded