What is this?
A plug-and-play Markdown skill file that forces AI agents to do real 'Deep Research' instead of shallow 'Wide Research' - instead of skimming many sources into a tidy checklist, it makes the agent closely read primary sources, surface contradictions, and land on a judgment that experts would actually argue about.
A plug-and-play Markdown skill file that forces AI agents to do real 'Deep Research' instead of shallow 'Wide Research' - instead of skimming many sources into a tidy checklist, it makes the agent closely read primary sources, surface contradictions, and land on a judgment that experts would actually argue about.Is this a full application or just a prompt file?It is a single Markdown skill file, not an app or a server - there is nothing to install, you just paste it into an AI agent's skill setup.
It is a single Markdown skill file, not an app or a server - there is nothing to install, you just paste it into an AI agent's skill setup.How do I actually use it?Copy the Chinese version (SKILL.md) or the English version (SKILL.en.md) into your agent's skill directory, or paste the file content directly as the agent's instructions.
Copy the Chinese version (SKILL.md) or the English version (SKILL.en.md) into your agent's skill directory, or paste the file content directly as the agent's instructions.Which AI agents and tools does it need?Any agent that can read a skill file and spawn sub-agents works - Claude Code, Cursor, and Cline are named examples. It also expects some kind of search/scraping capability, like a Tavily-style search MCP or the agent's built-in web search, and degrades gracefully depending on what is available.
Any agent that can read a skill file and spawn sub-agents works - Claude Code, Cursor, and Cline are named examples. It also expects some kind of search/scraping capability, like a Tavily-style search MCP or the agent's built-in web search, and degrades gracefully depending on what is available.What's the biggest risk when an agent runs this skill?Uncontrolled sub-agent recursion: a real incident escalated from 3 sub-agents to 13, then 50+, then 149, running up a surprise API bill - so the skill explicitly flags giving sub-agents their own search tool as a danger sign.
Uncontrolled sub-agent recursion: a real incident escalated from 3 sub-agents to 13, then 50+, then 149, running up a surprise API bill - so the skill explicitly flags giving sub-agents their own search tool as a danger sign.What are the four phases of the research workflow?Phase 0 forces the agent to question hidden assumptions before searching anything; Phase 1 is close reading of 3-5 primary sources; Phase 2 compares cases and deliberately hunts for failures and criticism; Phase 3 cross-validates agreeing versus conflicting sources; Phase 4 writes the argument in prose (not tables) and ends with a self-check that the conclusion would actually spark debate.
Phase 0 forces the agent to question hidden assumptions before searching anything; Phase 1 is close reading of 3-5 primary sources; Phase 2 compares cases and deliberately hunts for failures and criticism; Phase 3 cross-validates agreeing versus conflicting sources; Phase 4 writes the argument in prose (not tables) and ends with a self-check that the conclusion would actually spark debate.What common failure patterns does the skill guard against?It lists six specific ways agents cut corners - skipping Phase 0, letting sub-agent output become the report's structure, replacing analysis with tables and bullets, landing on 'safe' conclusions nobody would dispute, confusing breadth with depth, and letting sub-agents recursively spawn more sub-agents - each paired with a concrete detection signal.
It lists six specific ways agents cut corners - skipping Phase 0, letting sub-agent output become the report's structure, replacing analysis with tables and bullets, landing on 'safe' conclusions nobody would dispute, confusing breadth with depth, and letting sub-agents recursively spawn more sub-agents - each paired with a concrete detection signal.Why should I trust this skill's rules?The author says every anti-pattern in the file comes from a real research failure that was costly or embarrassing enough to turn into a rule, including the actual 149-agent runaway cost incident, rather than being theorized in the abstract.
The author says every anti-pattern in the file comes from a real research failure that was costly or embarrassing enough to turn into a rule, including the actual 149-agent runaway cost incident, rather than being theorized in the abstract.What are similar projects?ai-aug-edu, serenity-skill, zhangxuefeng-skill, skills, awesome-chatgpt-zh, shuohao-skills, seo-skills, hand-drawn-styles
ai-aug-edu, serenity-skill, zhangxuefeng-skill, skills, awesome-chatgpt-zh, shuohao-skills, seo-skills, hand-drawn-stylesWhich topics is it in?research-tools, deep-research, skill-md, mit-license, prompt, chinese-language, claude-code-skill, claude-skills, prompt-engineering, skill, cursor, agentic-ai
research-tools, deep-research, skill-md, mit-license, prompt, chinese-language, claude-code-skill, claude-skills, prompt-engineering, skill, cursor, agentic-ai