---

name: "search-retrieval" description: "Helps users choose Cognee search and recall modes, tune retrieval controls, and diagnose empty or invalid query behavior." disable-model-invocation: true metadata: disco-role: operating license: Apache 2.0

Cognee Search and Retrieval

Use this sub-skill when the user wants to query stored memory, compare retrieval modes, or understand why a search/recall query behaves the way it does.

Route here for

• search(...) and recall(...) mode selection.
• SearchType values and query routing.
• Session-aware retrieval, graph-only retrieval, code search, temporal search, and agentic completion.
• Query controls such as top_k, node_name, scope, datasets, dataset_ids, and include_references.
• Explaining empty results or invalid query combinations.

Route away

• Ingestion/storage/build workflows: core-memory.
• Provider and backend setup: configuration-backends.
• Session memory/feedback decorators and agent SDK flows: agent-session-memory.
• Custom graph schemas and pipelines: advanced-graphs-pipelines.
• CLI/service usage: api-cli-services.

Read first

1. references/search-modes.md
2. references/api-reference.md
3. references/troubleshooting.md

Safe helper

• scripts/choose_search_mode.py — suggests a search mode from intent flags and prints the rationale.

Working rules

• Use recall when the user wants session-aware retrieval or a memory-oriented phrasing.
• Use search when the user wants the lower-level graph query surface or explicit SearchType control.
• Always check whether the user asked for graph context, chunks, summaries, code traversal, or agentic completion; those intents map to different SearchType values.
• Validate code_query only with SearchType.CODE and skills/tools only with SearchType.AGENTIC_COMPLETION.
• Route provider/embedding/database problems back to the configuration sub-skill.