signaling-kb-builder
Build and maintain an entity-centric cell signaling knowledge base from lecture transcripts. Uses a three-stage LLM pipeline (Opus 4.5 extraction, Opus 4.5 semantic dedup, Haiku 4.5 intelligent merge) to create queryable intervention, pathway, biomarker, and conflict nodes.
npx skills add majiayu000/claude-skill-registry --skill signaling-kb-builder --agent claude-code
Same command for any agent — swap --agent for codex, cursor, copilot.
Weekly change comes from our own snapshots, not the repository page — it measures attention, not adoption.
What it does
Transforms lecture transcripts into an entity-centric knowledge graph optimized for clinical queries by extracting interventions, pathways, biomarkers, and conflicts, then merging into a structured JSON-driven knowledge base.
How it works
- Uses a three-stage pipeline: EXTRACTION (Claude Opus 4.5) to identify interventions, pathways, biomarkers, and facts with citations and confidence; SEMANTIC DEDUP (Claude Opus 4.5) for mapping entities to existing KB entries and alias resolution; MERGE (Haiku 4.5) for intelligent deduplication and integration into existing JSON nodes.
- Output is organized under .signaling-kb with subfolders interventions, pathways, biomarkers, conflicts, sources, state, and exports.
- Each dosing protocol and mechanism claim is enriched with up to five peer-reviewed citations via a multi-step enrichment process (Perplexity Deep Research, URL parsing, Haiku fallback) and then merged into the node structure.
- Maintains processing state and a cost-report for enrichment activities; commits are written to disk as JSON.
When to use it
- When you want to convert transcripts into a queryable, entity-centric KB for intervention dosing, pathway mechanisms, biomarker interpretation, and related conflicts.
- For long runs, it supports background execution and monitoring via the provided scripts.
What it can touch
- Reads transcripts from Google Drive (sync_gdrive_transcripts.py), performs extraction (extract_kb_openrouter.py), and stores results in the .signaling-kb directory.
- Creates and updates these node types: interventions, pathways, biomarkers, conflicts, plus sources and exports. Enrichment pulls in peer-reviewed citations and stores them with claims.
Caveats
- Conflict detection is not implemented yet (status indicated as aspirational; currently appends all claims).
- Aliases are tracked but may be stored as separate files due to slugification rules; future enhancements include LLM-based semantic entity resolution.
- Enrichment has cost implications disclosed in the pipeline description.
# Cell Signaling Knowledge Base Builder Transform lecture transcripts into an entity-centric knowledge graph optimized for clinical queries (biomarker interpretation, intervention dosing, pathway mechanisms). ## Why This Skill Exists ### Problem Statement The existing VectorShift knowledge base stores lecture analyses as document chunks. When queried: | Query Type | Current Behavior | Desired Behavior | |------------|------------------|------------------| | "rapamycin dosing" | Returns chunks mentioning mTOR (wrong compound) | Returns all rapamycin dosing protocols | | "low IGF-1 with insulin resistance" | Returns how to RAISE IGF-1 | Returns diagnostic INTERPRETATION | | "thymosin alpha-1 dosing" | Returns unrelated peptides (TB-500, BPC-157) | Returns TA1 protocols (content exists!) | | "AMPK autophagy" | Works well (vocabulary matches) | Works well | **Root cause:** Semantic search matches entity mentions but doesn't understand clinical query intent. Document-centric chunking scatters related information. ### Solution Reorganize knowledge from **lecture-centric** to **entity-centric**: ``` BEFORE: Lectures → Chunks → Semantic Search AFTER: Lectures → Extract → Entity Nodes → Int
- Why This Skill Exists
- Problem Statement
- Solution
- Architecture Overview (Current Implementation)
- Quick Commands
- Primary Workflow (Recommended)
- View Results
- Future Commands (Not Yet Implemented)
- Output Structure
- Node Schemas
- Intervention Node
- Pathway Node
- Biomarker Interpretation Node
- Processing Pipeline (Detailed)
python3 scripts/sync_gdrive_transcripts.py \ python3 scripts/extract_kb_openrouter.py --status nohup python3 scripts/extract_kb_openrouter.py --from-tracker > extraction.log 2>&1 & nohup bash scripts/monitor_extraction.sh > monitor.log 2>&1 & Count all entity files ls .signaling-kb/interventions/*.json | wc -l # ~580 files ls .signaling-kb/pathways/*.json | wc -l # ~207 files ls .signaling-kb/biomarkers/*.json | wc -l # ~239 files ls .signaling-kb/conflicts/*.json | wc -l # ~576 files View specific entity
What does the signaling-kb-builder skill do?
Build and maintain an entity-centric cell signaling knowledge base from lecture transcripts. Uses a three-stage LLM pipeline (Opus 4.5 extraction, Opus 4.5 semantic dedup, Haiku 4.5 intelligent merge) to create queryable intervention, pathway, biomarker, and conflict nodes.
How do I install it?
Run `npx skills add majiayu000/claude-skill-registry --skill signaling-kb-builder --agent claude-code` — it drops the skill into your project so the agent can pick it up. Swap the --agent value for codex, cursor or copilot if you use one of those.
Where does this skill come from?
From majiayu000/claude-skill-registry, a repository with 534 stars. We read it straight from the repository tree rather than a submitted listing, so what you see here is what is actually published.
Is a popular skill a good skill?
Not necessarily. Stars measure attention, not adoption — a repository can trend for a week and be abandoned. That is why we show the weekly change from our own snapshots next to the total, instead of a single flattering number.