From deep-research-core
Recherche autonome local-first : collecte multi-sources, deduplication, scoring et synthese Markdown. Declencheurs : /tk:deep-research, recherche approfondie, synthese de sources, pipeline de recherche, deep research.
How this skill is triggered — by the user, by Claude, or both
Slash command
/deep-research-core:tk-deep-researchThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
> Local-first autonomous research: collect, deduplicate, score and synthesize.
Local-first autonomous research: collect, deduplicate, score and synthesize.
/tk:deep-research "<query>"
/tk:deep-research --pipeline content "<query>"
/tk:deep-research --pipeline github "<query>"
/tk:deep-research --dry-run "<query>"
1. Identify the pipeline based on the query (content / entity / github / vendor)
2. Read sources/trusted_sources.yml → select relevant sources
3. Block any source in blocked_sources.yml
4. Collect in parallel (max 3 simultaneous sources)
5. Deduplicate by title hash + semantic similarity (threshold 0.85)
6. Score each result (0.0 → 1.0) according to scoring_weights
7. Filter: keep score >= 0.70 only
8. Synthesize in structured Markdown
9. Index in Obsidian (target folder per pipeline)
10. Index in Qdrant if available
11. Return final report with cited sources
trusted_sources.yml.planning/DECISIONS.md if research leads to a decision| Pipeline | Usage |
|---|---|
| content | Domain content watch, entity profiles |
| entity | Authors, publishers, producers |
| github | Repo audits, integration scoring |
| vendor | VPS, services, comparisons |
{
"status": "success",
"skill_name": "tk-deep-research",
"query": "<query>",
"pipeline": "<pipeline>",
"items_found": N,
"items_after_dedup": N,
"items_above_threshold": N,
"report_path": "vault/reports/research_YYYY-MM-DD_<slug>.md",
"output": {
"summary": "...",
"data": [],
"sources_used": []
}
}
vault/reports/ locallyCreates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.
npx claudepluginhub geekfamilycorp/tricorderkit --plugin deep-research-core