topic-research-deposition

name	topic-research-deposition
description	Use when the user asks to 搜一下、调研、找资料、沉淀素材、补截图、做产品事实收集、投资/OSS research、公众号选题搜索、写作前第 0 步，或需要在写作/产品分析前建立可复查的 evidence folder.

Topic Research Deposition

This skill collects source material before writing, product analysis, investment review, or general research. It deposits evidence; it does not write the final article, product verdict, or investment memo.

Core Rule

Default search and access tool: agent-reach.

Twitter/X, Reddit, Exa/web, 微信公众号, GitHub, Hacker News, YouTube, podcasts, papers, and other platform searches are treated as agent-reach reach surfaces. Do not turn them into a required sequence. Use direct platform commands only as fallback when agent-reach is unavailable, incomplete, or the user explicitly asks for a direct platform command.

公众号账号级历史、批量 URL 和对标账号语料由 wechat-account-corpus-research 负责。用户明确要求这些能力时，先调用该瓦技能，再消费其 manifest；本技能不要平行实现第二套账号抓取。

SKILL.md is the router. Load only the references required by the selected mode. Do not load all WeChat, product, and fallback command references together.

Mode Selection

Choose one mode before creating folders or searching.

Mode	Use When	Required References
general-research	General research, background checks, source collection, concept definition, broad web/social search	references/agent-reach-search.md, references/quality-checklist.md
writing-research	Research for an article that is not necessarily WeChat-specific	references/agent-reach-search.md, references/quality-checklist.md
wechat-writing-research	公众号选题、第 0 步、需要微信素材、公众号爆款逻辑、图文截图和后续交给 public-account-writing-style	references/wechat-writing-research.md, references/wechat-viral-logic.md, references/wechat-extraction.md, references/quality-checklist.md
product-research	Facts needed before ai-product-analyzer: product, company, pricing, customers, competitors, demo, docs, GitHub, traction	references/product-investment-research.md, references/agent-reach-search.md, references/quality-checklist.md
investment-research	Facts needed before product-analysis structured output: financing, team, market, OSS metrics, commercial traction, DD evidence	references/product-investment-research.md, references/agent-reach-search.md, references/quality-checklist.md

If the user says "公众号", "第 0 步", "爆款", "推草稿箱前", or asks for WeChat screenshots, choose wechat-writing-research.

If the user asks for product analysis, product visual report, investment report, or OSS investment and facts are missing, choose product-research or investment-research before analysis.

Standard Flow

clarify research objective
-> choose mode
-> create mode-appropriate research folder
-> [if account history or benchmark corpus is requested] consume wechat-account-corpus-research output
-> search with agent-reach
-> save raw source material by source/platform
-> screenshot or archive evidence according to the selected mode
-> quality check
-> report coverage and stop for next instruction

Output Contract

Every run must end with:

Mode:
Research folder:
Search tool:
Sources collected:
Screenshots / archived evidence:
Coverage:
Gaps:
Next suggested handoff:

Handoff rules:

• writing-research and wechat-writing-research can hand off to public-account-writing-style.
• 交接后由 public-account-writing-style 先完成编辑判断，再决定是否使用框架并完成起草与终审；本技能只提供研究过程包。
• product-research can hand off to ai-product-analyzer.
• investment-research can hand off to workflow-orchestrator product-analysis chain for OSS investment structured and visual output.
• 独立调用本技能时，完成 coverage report 后停止，等待用户决定是否继续。
• 用户已明确要求运行完整 wechat-writing chain 时，由 chain 管理交接；只读研究门禁通过后不额外暂停，但新增登录、扩范围或发布授权仍必须单独确认。

Storage Rules

• General/product/investment research should use a neutral research folder, not the WeChat run folder.
• Only wechat-writing-research uses writing/drafts/{YYYY-MM-DD}-{topic-slug}/; its source material and screenshots live under 01-topic-research/.
• Keep raw source material separate from analysis. Do not turn the deposition folder into a finished article or memo.
• Save one source per file when practical. Include title, author/source, date, URL, capture method, search mode, and full text or full relevant excerpt.
• If a source cannot be fully fetched, mark it as partial and explain why. Do not hide the failure behind a summary.

Common Mistakes

Mistake	Correction
Treating every research task as WeChat writing	Choose mode first; WeChat is only one mode.
Treating a platform list as a required search sequence	Use agent-reach as the default reach layer; platforms are surfaces, not workflow nodes.
Losing existing WeChat experience while generalizing	Load wechat-writing-research.md and wechat-viral-logic.md when mode is WeChat.
Summarizing sources instead of depositing them	Save raw source material first; analysis belongs to the next skill.
Rebuilding account-history crawling inside this skill	Call wechat-account-corpus-research, then consume its manifest and corpus.
Continuing after a standalone research call	Stop after coverage report; only an explicitly requested end-to-end chain may continue through its declared handoff.
Claiming agent-reach was used when it was not available	State the actual fallback command or tool used.

Reference Files

File	When to Read
references/agent-reach-search.md	Default search behavior, source surfaces, fallback policy
references/wechat-writing-research.md	WeChat-specific folder layout, confirmation questions, rounds, screenshot and report format
references/wechat-viral-logic.md	WeChat virality and search-to-writing logic; required for wechat-writing-research
references/wechat-extraction.md	WeChat fetch/CAPTCHA fallback details
references/product-investment-research.md	Product, company, OSS, investment fact collection
references/platform-commands.md	Direct command fallback only when agent-reach cannot cover a platform
references/quality-checklist.md	Final quality check for source completeness and deposition discipline

完成标准

• 完成本 Skill 正文中 Mode、Output、Handoff 或最终交付步骤声明的全部适用产物，不因它处在较大任务中而缩减。
• 执行正文已有的检查清单、视觉检查、平台兼容检查或事实检查；不适用项说明理由，失败项先在本 Skill 内返工。
• 返回用户可直接使用或交给下一个完整 Skill 的独立产物，并明确仍需用户授权的外部写入。