Summary

A diagnostic skill for identifying and resolving Qdrant vector search quality issues. It guides agents to analyze embedding models, Qdrant configuration, and query strategies—distinguishing configuration problems from upstream data/model problems. The skill is deliberately minimal, serving as an index to sub-skills rather than standalone instructions.

Detected Capabilities

Trigger Keywords

Phrases that MCP clients use to match this skill to user intent.

Risk Signals

Referenced Domains

External domains referenced in skill content, detected by static analysis.

Use Cases

• Diagnose why semantic search returns irrelevant results
• Improve search precision and recall after data growth
• Choose between embedding models based on quality metrics
• Decide whether to implement hybrid search or reranking
• Troubleshoot search quality degradation after quantization or model updates

Quality Notes

• Strengths: Clear problem detection heuristics (e.g., 'splitting mid-sentence can drop quality 30-40%') ground diagnosis in real data. Allowed-tools field correctly limits scope to read-only operations (Read, Grep, Glob).
• Weakness: Skill is primarily an index/stub referencing external sub-skills that are not included in the delivery. An agent reading this skill cannot execute diagnosis without external files.
• Weakness: No concrete diagnostic steps, parameters to inspect, or examples. Instructions jump to 'test with exact search' and 'tune HNSW parameters' without explaining what to measure or how.
• Weakness: No edge cases, error handling, or validation guidance. What if exact search also fails? How to distinguish embedding model quality from chunking quality?
• Improvement opportunity: Inline or embed key diagnostic workflows rather than delegating entirely to missing sub-skills.