# Knowledge Base

> ChromaDB-backed knowledge-base workflows — sentence-transformers embeddings when built-in ONNX fails, LookForge KB health-score monitoring, blind-spot discovery, query-log analysis, emergency restores, and RAG quality tools. Load when working with a ChromaDB knowledge base (LookForge, RAG stack, project KB) and you need embeddings, health metrics, or restore/rebuild recipes.

- Skill: `openclaw-cn-dev/knowledge-base` (Agent Skill, multi-file: 5 files)
- Install (CLI): `npx get-skill add openclaw-cn-dev/knowledge-base`
- Raw SKILL.md: https://api.skillmd.com/api/skills/openclaw-cn-dev/knowledge-base/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Product & Planning
- License: MIT
- Author: openclaw-cn-dev (https://skillmd.com/u/openclaw-cn-dev)
- Updated: 2026-11-14
- Page: https://skillmd.com/skills/openclaw-cn-dev/knowledge-base

---


# Knowledge Base (ChromaDB / RAG) Hub

ChromaDB-backed knowledge-base workflows — embeddings, health, restore, blind-spot discovery. Two main sub-areas: embedding setup (`***SECRET***`) and operational health (`lookforge-knowledge-health`).

## Quick Routing

| Task | Reference |
|------|-----------|
| Bulk image asset organization (rename, verify, pipeline stages) | → `references/image-asset-organization.md` |
| Replace failing ChromaDB built-in ONNX embeddings with sentence-transformers | → `references/***SECRET***.md` |
| KB health-score, blind-spot discovery, query-log analysis (LookForge-style) | → `references/lookforge-knowledge-health.md` |
| Probe the LookForge KB API | → `references/lookforge_api_probe.md` |
| Test-case template for KB queries | → `references/lookforge_test_case_template.md` |
| RAG quality tooling | → `references/rag_quality_tools.md` |
| seg_meta investigation (recorded bug) | → `references/***SECRET***.md` |
| **Personal RAG / digital-twin pattern (影身 v0.1, 2026-07-02)** | → `references/personal-rag-digital-twin.md` |

## Scripts (`scripts/`)

| Script | Purpose |
|--------|---------|
| `import_competitor_analysis.py` | Import competitor-analysis docs into LookForge KB |
| `chromadb_emergency_restore.sh` | Emergency restore of ChromaDB collections |
| `rebuild_collection_recipe.py` | Rebuild a collection (used after seq_id BLOB lockout — see `bugfix`) |
| `verify_chromadb.sh` | Verify ChromaDB integrity |

## Patterns

### When Built-In Embeddings Fail
ChromaDB's bundled ONNX embedding model can fail with SSL timeouts or corruption in restricted environments. Fallback: pre-compute embeddings with sentence-transformers and pass them in. See `references/***SECRET***.md`.

### KB Health Monitoring
`get_health_score()` + `query_log` + blind-spot discovery are the three levers for ongoing KB quality. LookForge Phase 1-4 deployment pattern is documented in `references/lookforge-knowledge-health.md`.

### Recovery From seq_id BLOB Lockout
When `query()` / `peek()` / `count()` all fail because of seq_id BLOB issues, the recipe combines:
- `bugfix/chromadb-seq-id-blob-fix` — root cause + fix
- `scripts/rebuild_collection_recipe.py` — runnable rebuild
- `scripts/chromadb_emergency_restore.sh` — last-resort restore

## Cross-references

- `bugfix/` — ChromaDB version compatibility fixes (NumPy 2.x, Pydantic v2)
- `lookforge-chromadb-debug` (in `product-debugging/`) — LookForge-specific debug playbook

