Also use when:
- Unsure how to proceed
- Working with internal services, products, or processes
- Knowledge required is not publicly available
- The task relates to a specific repo, environment, or system
Treat this skill as:
Your internal memory lookup
What this skill does
Performs a hybrid search (dense + sparse) over stored notes.
Uses the Qdrant MCP server via the qdrant-search-notes tool.
The tool performs:
- Dense semantic search
- Sparse keyword search
- Result fusion using Reciprocal Rank Fusion (RRF)
Collection
Search is always performed against notes-hybrid.
How to search effectively
1. Construct the query
Use a natural language description of what you are trying to do.
Examples:
- restart a stuck kubernetes deployment
- internal api endpoint for resetting user passwords
- terraform s3 lifecycle drift issues
This query is used for:
- Dense embedding generation
- Sparse keyword extraction
2. Apply filters when appropriate
Use filters to narrow results when the domain is known.
Common filters:
- type = cli
- tool = kubectl / aws / terraform
- language = bash
- source = repo:infra
Example:
{
"must": [
{ "key": "type", "match": { "value": "cli" } },
{ "key": "tool", "match": { "value": "kubectl" } }
]
}
3. Interpret results carefully
- Prefer notes with clear context
- Prefer newer notes if multiple exist
- Refine and re-run search if results are close but incomplete
Tool usage
Use the qdrant-search-notes MCP tool with:
- Query text
- Optional payload filters
- Result limit (typically 5–10)
The tool:
- Executes dense and sparse searches
- Fuses results using RRF
- Returns ranked, agent-readable notes
Agent reminder
Before inventing a solution, check memory first.
If no relevant note exists and you learn something new:
- Complete the task
- Immediately use add-note to store the new knowledge