[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M72SW7CK2VBH4EK7J8AHGE80":3,"skill-same:01M72SW7CK2VBH4EK7J8AHGE80":65},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":19,"linkOnly":20,"markdownContent":21,"markdownBody":22,"frontMatter":23,"refName":44,"commitSha":45,"skillDirectory":46,"forksCount":47,"repositoryHtmlUrl":48,"repositoryHomepageUrl":6,"ownerHtmlUrl":49,"ownerBlogUrl":6,"ownerDescription":50,"updatedAt":51,"install":52,"resources":64},"01M72SW7CK2VBH4EK7J8AHGE80","data-research",null,"Structured data research: search sources, extract structured data,\narchive raw sources, maintain canonical tracker pages, deduplicate.\nParameterized via YAML recipes for investor updates, donations,\ncompany updates, or any email-to-structured-data pipeline.","ltdrew\u002FAgent-brain-2","ltdrew","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F173687628?v=4","01M72SVABERGYY99JQPNV3FTTN","Logan Drew","01M72SVABZHAN29CHRV66Y48B8",1,"MIT","https:\u002F\u002Fgithub.com\u002Fltdrew\u002FAgent-brain-2\u002Fblob\u002Ffeat\u002Fbrain-first-convention\u002Fskills\u002Fdata-research\u002FSKILL.md",[],[],"2026-11-10 22:39:42",false,"---\nname: data-research\nversion: 1.0.0\ndescription: |\n  Structured data research: search sources, extract structured data,\n  archive raw sources, maintain canonical tracker pages, deduplicate.\n  Parameterized via YAML recipes for investor updates, donations,\n  company updates, or any email-to-structured-data pipeline.\ntriggers:\n  - \"research\"\n  - \"track\"\n  - \"extract from email\"\n  - \"investor updates\"\n  - \"donations\"\n  - \"build a tracker\"\n  - \"data dig\"\ntools:\n  - search\n  - query\n  - get_page\n  - put_page\n  - add_link\n  - add_timeline_entry\n  - put_raw_data\n  - file_upload\nmutating: true\n---\n\n# Data Research\n\nStructured research pipeline: search sources, extract structured data,\narchive raw, deduplicate, update canonical trackers, backlink entities.\n\n## Contract\n\nOne skill for any email-to-structured-data pipeline. The only differences\nbetween tracking investor updates, expenses, and company metrics\nare the **search queries**, **extraction schemas**, and **tracker page format**.\nAll three use the same 7-phase pipeline with parameterized recipes.\n\n## When to Use\n\n- User wants to track structured data from email, web, or API sources\n- User says \"research\", \"track\", \"extract from email\", \"build a tracker\"\n- User mentions investor updates, donations, company metrics, filings\n- User wants to set up recurring data collection (with cron recipe)\n\n## Phases\n\n### Phase 1: Define Research Recipe\n\nAsk the user what they want to track. Either:\n- Pick a built-in recipe: investor-updates, expense-tracker, company-updates\n- Define a custom recipe with: source queries, classification rules, extraction schema,\n  tracker page path, tracker format\n\nRecipes are YAML files at `~\u002F.gbrain\u002Frecipes\u002F{name}.yaml`. Use `gbrain research init`\nto scaffold a new one.\n\n### Phase 2: Search Sources\n\nBrain first (maybe we already have this data). Then:\n- **Email** via credential gateway: windowed queries (quarterly, monthly if truncated)\n- **Web** via search: public filings, press releases, regulatory data\n- **APIs**: any structured data source the recipe defines\n- **Attachments**: PDF extraction, HTML stripping\n\n### Phase 3: Classify\n\nDeterministic first (regex patterns from recipe), LLM fallback.\nLog every LLM fallback for future regex improvement (fail-improve loop).\nSkip marketing, newsletters, noise based on recipe's classification rules.\n\n### Phase 4: Extract Structured Data\n\n**EXTRACTION INTEGRITY RULE:**\n1. Save raw source immediately (before any extraction)\n2. Extract fields using deterministic regex first, LLM fallback\n3. When summarizing batch results: **re-read from saved files**\n4. Never trust LLM working memory after batch processing\n\nThis prevents a known hallucination bug where batch-processed amounts were\n13\u002F13 wrong from LLM working memory while saved files were correct.\n\n### Phase 5: Archive Raw Sources\n\n- `put_raw_data` for email bodies, API responses\n- `file_upload` for PDF attachments, documents\n- Create `.redirect.yaml` pointers for large files in storage\n- Every tracker entry must link back to its raw source\n\n### Phase 6: Deduplicate\n\nBefore adding to tracker:\n- Exact match (same key fields) → skip\n- Fuzzy match (same entity + date + similar amount within tolerance) → flag for review\n- Different amount for same entity+date → add with note (could be correction)\n\n### Phase 7: Update Canonical Tracker + Backlink\n\n- Parse existing tracker page (markdown table)\n- Append new entries in correct section (grouped by year\u002Fquarter\u002Fentity)\n- Compute running totals\n- Backlink every mentioned entity (person → people\u002F page, company → companies\u002F page)\n- Uses enrichment service for entity pages\n\n## Built-In Recipes\n\nThree example recipes ship with GBrain (see `~\u002F.gbrain\u002Frecipes\u002F`):\n\n1. **investor-updates** — extract MRR, ARR, growth, burn, runway, headcount from investor update emails\n2. **expense-tracker** — extract amounts, recipients, platforms from receipt emails (subscriptions, services, recurring charges)\n3. **company-updates** — extract revenue, users, key metrics from portfolio company update emails\n\n## Anti-Patterns\n\n- Trusting LLM working memory for amounts after batch processing (use extraction integrity rule)\n- Creating tracker entries without raw source links\n- Running without deduplication (leads to double-counted entries)\n- Hardcoding source-specific patterns in the pipeline code (use recipes)\n\n## Output Format\n\nBrain page at the recipe's `tracker_page` path with markdown tables:\n\n```markdown\n### 2026\n\n| Date | Company | MRR | ARR | Growth | Status |\n|------|---------|-----|-----|--------|--------|\n| 2026-04-01 | Example Co | $188K | $2.3M | +14.7% MoM | [Source](link) |\n```\n\nEach entry links to its raw source. Running totals at the bottom of each section.\n\n## Conventions\n\nReferences `skills\u002Fconventions\u002Fquality.md` for citation and back-linking rules.\n","# Data Research\n\nStructured research pipeline: search sources, extract structured data,\narchive raw, deduplicate, update canonical trackers, backlink entities.\n\n## Contract\n\nOne skill for any email-to-structured-data pipeline. The only differences\nbetween tracking investor updates, expenses, and company metrics\nare the **search queries**, **extraction schemas**, and **tracker page format**.\nAll three use the same 7-phase pipeline with parameterized recipes.\n\n## When to Use\n\n- User wants to track structured data from email, web, or API sources\n- User says \"research\", \"track\", \"extract from email\", \"build a tracker\"\n- User mentions investor updates, donations, company metrics, filings\n- User wants to set up recurring data collection (with cron recipe)\n\n## Phases\n\n### Phase 1: Define Research Recipe\n\nAsk the user what they want to track. Either:\n- Pick a built-in recipe: investor-updates, expense-tracker, company-updates\n- Define a custom recipe with: source queries, classification rules, extraction schema,\n  tracker page path, tracker format\n\nRecipes are YAML files at `~\u002F.gbrain\u002Frecipes\u002F{name}.yaml`. Use `gbrain research init`\nto scaffold a new one.\n\n### Phase 2: Search Sources\n\nBrain first (maybe we already have this data). Then:\n- **Email** via credential gateway: windowed queries (quarterly, monthly if truncated)\n- **Web** via search: public filings, press releases, regulatory data\n- **APIs**: any structured data source the recipe defines\n- **Attachments**: PDF extraction, HTML stripping\n\n### Phase 3: Classify\n\nDeterministic first (regex patterns from recipe), LLM fallback.\nLog every LLM fallback for future regex improvement (fail-improve loop).\nSkip marketing, newsletters, noise based on recipe's classification rules.\n\n### Phase 4: Extract Structured Data\n\n**EXTRACTION INTEGRITY RULE:**\n1. Save raw source immediately (before any extraction)\n2. Extract fields using deterministic regex first, LLM fallback\n3. When summarizing batch results: **re-read from saved files**\n4. Never trust LLM working memory after batch processing\n\nThis prevents a known hallucination bug where batch-processed amounts were\n13\u002F13 wrong from LLM working memory while saved files were correct.\n\n### Phase 5: Archive Raw Sources\n\n- `put_raw_data` for email bodies, API responses\n- `file_upload` for PDF attachments, documents\n- Create `.redirect.yaml` pointers for large files in storage\n- Every tracker entry must link back to its raw source\n\n### Phase 6: Deduplicate\n\nBefore adding to tracker:\n- Exact match (same key fields) → skip\n- Fuzzy match (same entity + date + similar amount within tolerance) → flag for review\n- Different amount for same entity+date → add with note (could be correction)\n\n### Phase 7: Update Canonical Tracker + Backlink\n\n- Parse existing tracker page (markdown table)\n- Append new entries in correct section (grouped by year\u002Fquarter\u002Fentity)\n- Compute running totals\n- Backlink every mentioned entity (person → people\u002F page, company → companies\u002F page)\n- Uses enrichment service for entity pages\n\n## Built-In Recipes\n\nThree example recipes ship with GBrain (see `~\u002F.gbrain\u002Frecipes\u002F`):\n\n1. **investor-updates** — extract MRR, ARR, growth, burn, runway, headcount from investor update emails\n2. **expense-tracker** — extract amounts, recipients, platforms from receipt emails (subscriptions, services, recurring charges)\n3. **company-updates** — extract revenue, users, key metrics from portfolio company update emails\n\n## Anti-Patterns\n\n- Trusting LLM working memory for amounts after batch processing (use extraction integrity rule)\n- Creating tracker entries without raw source links\n- Running without deduplication (leads to double-counted entries)\n- Hardcoding source-specific patterns in the pipeline code (use recipes)\n\n## Output Format\n\nBrain page at the recipe's `tracker_page` path with markdown tables:\n\n```markdown\n### 2026\n\n| Date | Company | MRR | ARR | Growth | Status |\n|------|---------|-----|-----|--------|--------|\n| 2026-04-01 | Example Co | $188K | $2.3M | +14.7% MoM | [Source](link) |\n```\n\nEach entry links to its raw source. Running totals at the bottom of each section.\n\n## Conventions\n\nReferences `skills\u002Fconventions\u002Fquality.md` for citation and back-linking rules.\n",{"name":5,"tools":24,"version":33,"mutating":34,"triggers":35,"description":43},[25,26,27,28,29,30,31,32],"search","query","get_page","put_page","add_link","add_timeline_entry","put_raw_data","file_upload","1.0.0",true,[36,37,38,39,40,41,42],"research","track","extract from email","investor updates","donations","build a tracker","data dig","Structured data research: search sources, extract structured data,\narchive raw sources, maintain canonical tracker pages, deduplicate.\nParameterized via YAML recipes for investor updates, donations,\ncompany updates, or any email-to-structured-data pipeline.\n","feat\u002Fbrain-first-convention","f3dace916c3845569f26904591b95baf00237bf8","skills\u002Fdata-research",0,"https:\u002F\u002Fgithub.com\u002Fltdrew\u002FAgent-brain-2","https:\u002F\u002Fgithub.com\u002Fltdrew","Account 2","2026-05-13 07:52:42",{"source":8,"skillName":5,"commands":53,"prompt":63,"zipAvailable":34},[54,57,60],{"tool":55,"command":56},"npx","npx skills add ltdrew\u002FAgent-brain-2 --skill data-research",{"tool":58,"command":59},"bunx","bunx skills add ltdrew\u002FAgent-brain-2 --skill data-research",{"tool":61,"command":62},"pnpm","pnpm dlx skills add ltdrew\u002FAgent-brain-2 --skill data-research","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fltdrew\u002FAgent-brain-2\u002Fblob\u002Ffeat\u002Fbrain-first-convention\u002Fskills\u002Fdata-research\u002FSKILL.md",[],[66,73,79,86,93,100],{"skillNo":67,"name":68,"nameZh":6,"description":69,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":70,"categories":71,"keywords":72,"publishedAt":19},"01M72SYQK7HQ98AEY8Z8DBT3C0","signal-detector","Always-on ambient signal capture for the OpenClaw reference fixture.","https:\u002F\u002Fgithub.com\u002Fltdrew\u002FAgent-brain-2\u002Fblob\u002Ffeat\u002Fbrain-first-convention\u002Ftest\u002Ffixtures\u002Fopenclaw-reference-minimal\u002Fskills\u002Fsignal-detector\u002FSKILL.md",[],[],{"skillNo":74,"name":26,"nameZh":6,"description":75,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":76,"categories":77,"keywords":78,"publishedAt":19},"01M72SYPHM4TDK6ZRV4NWAGC3D","Look up brain pages in the OpenClaw reference fixture.","https:\u002F\u002Fgithub.com\u002Fltdrew\u002FAgent-brain-2\u002Fblob\u002Ffeat\u002Fbrain-first-convention\u002Ftest\u002Ffixtures\u002Fopenclaw-reference-minimal\u002Fskills\u002Fquery\u002FSKILL.md",[],[],{"skillNo":80,"name":81,"nameZh":6,"description":82,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":83,"categories":84,"keywords":85,"publishedAt":19},"01M72SYN2J4BGSFS613D80AJQX","context-now","ALWAYS-ON time-sensitivity discipline for the OpenClaw reference fixture.","https:\u002F\u002Fgithub.com\u002Fltdrew\u002FAgent-brain-2\u002Fblob\u002Ffeat\u002Fbrain-first-convention\u002Ftest\u002Ffixtures\u002Fopenclaw-reference-minimal\u002Fskills\u002Fcontext-now\u002FSKILL.md",[],[],{"skillNo":87,"name":88,"nameZh":6,"description":89,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":90,"categories":91,"keywords":92,"publishedAt":19},"01M72SYKY67RCNH8VH9R2N0ZZV","brain-ops","Core read\u002Fwrite cycle for the OpenClaw reference fixture.","https:\u002F\u002Fgithub.com\u002Fltdrew\u002FAgent-brain-2\u002Fblob\u002Ffeat\u002Fbrain-first-convention\u002Ftest\u002Ffixtures\u002Fopenclaw-reference-minimal\u002Fskills\u002Fbrain-ops\u002FSKILL.md",[],[],{"skillNo":94,"name":95,"nameZh":6,"description":96,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":97,"categories":98,"keywords":99,"publishedAt":19},"01M72SYHEMGTDRWSERJWZR0XE2","webhook-transforms","Generic framework for converting external events (SMS, meetings, social mentions)\ninto brain-ingestible signals. Define a transform function, register a webhook URL,\nand incoming events get processed through the brain pipeline.","https:\u002F\u002Fgithub.com\u002Fltdrew\u002FAgent-brain-2\u002Fblob\u002Ffeat\u002Fbrain-first-convention\u002Fskills\u002Fwebhook-transforms\u002FSKILL.md",[],[],{"skillNo":101,"name":102,"nameZh":6,"description":103,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":104,"categories":105,"keywords":106,"publishedAt":19},"01M72SYEE2B8F47TN01CAGQ4JG","testing","Skill validation framework. Validates every skill has SKILL.md with frontmatter,\nevery reference exists, every env var is declared. The testing contract for the\nskill system itself.","https:\u002F\u002Fgithub.com\u002Fltdrew\u002FAgent-brain-2\u002Fblob\u002Ffeat\u002Fbrain-first-convention\u002Fskills\u002Ftesting\u002FSKILL.md",[],[]]