{
  "skill": {
    "id": "internal-deep-research",
    "name": "internal-deep-research",
    "description": "Use when you need to research internal materials from Nextcloud or Feishu, reconcile evidence across PDFs, DOCX, MD, and TXT files, and deliver a structured report.",
    "path": "skills/research/internal-deep-research/SKILL.md",
    "tags": [
      "research",
      "nextcloud",
      "feishu",
      "evidence",
      "reporting"
    ]
  },
  "document": {
    "path": "skills/research/internal-deep-research/SKILL.md",
    "frontmatter": {
      "name": "internal-deep-research",
      "description": "Use when you need to research internal materials from Nextcloud or Feishu, reconcile evidence across PDFs, DOCX, MD, and TXT files, and deliver a structured report.",
      "version": "0.1.0",
      "author": "Hermes Agent",
      "license": "private",
      "metadata": {
        "hermes": {
          "tags": [
            "research",
            "nextcloud",
            "feishu",
            "evidence",
            "reporting"
          ],
          "related_skills": [
            "feishu-doc-reporting",
            "api-debugging"
          ]
        }
      }
    },
    "content": "---\nname: internal-deep-research\ndescription: Use when you need to research internal materials from Nextcloud or Feishu, reconcile evidence across PDFs, DOCX, MD, and TXT files, and deliver a structured report.\nversion: 0.1.0\nauthor: Hermes Agent\nlicense: private\nmetadata:\n  hermes:\n    tags: [research, nextcloud, feishu, evidence, reporting]\n    related_skills: [feishu-doc-reporting, api-debugging]\n---\n\n# Internal Deep Research\n\n## Overview\n\nUse this skill when a user asks for a research-grade synthesis based on internal documents. The goal is not just to summarize text, but to collect evidence from source documents, compare claims, identify conflicts, and produce a concise report with traceable sources.\n\nThis skill is optimized for private knowledge workflows where the source of truth lives in Nextcloud files, Feishu cloud docs, or exported local documents.\n\n## When to Use\n\n- The user wants a Deep Research style report from internal materials\n- The task spans multiple documents or file types\n- Evidence needs to be cited or cross-checked\n- The final output should become a Feishu document\n- You need to separate facts, inferences, and recommendations\n\n## Workflow\n\n1. Clarify the research question and desired output format.\n2. Search relevant files by name and content.\n3. Read the most relevant PDFs, DOCX, MD, and TXT sources.\n4. Extract claims, numbers, dates, names, and decisions.\n5. Cross-check overlapping sources for consistency.\n6. Produce a structured synthesis with:\n   - executive summary\n   - key findings\n   - evidence table\n   - uncertainties and contradictions\n   - recommended next steps\n7. Create or update a Feishu document as the final artifact.\n\n## Output Standard\n\nA strong report should include:\n\n- Short executive summary\n- Bullet list of key findings\n- Source references with file names or doc links\n- Clear distinction between evidence and interpretation\n- Actionable conclusion\n\n## Common Pitfalls\n\n1. **Summarizing without evidence.** Always tie conclusions back to files or doc sections.\n2. **Mixing facts with recommendations.** Keep them separated.\n3. **Ignoring format differences.** PDFs, DOCX, and Feishu docs often need different extraction approaches.\n4. **Overloading the final report.** Keep the executive summary short and push detail into sections.\n5. **Skipping the deliverable.** The final report should end up in Feishu when requested.\n\n## Verification Checklist\n\n- [ ] Relevant files were searched by name and content\n- [ ] Primary sources were read, not just metadata\n- [ ] Claims were cross-checked across multiple sources when possible\n- [ ] The report distinguishes evidence from inference\n- [ ] A Feishu document was created or updated when requested\n"
  }
}