| agents | ["system-architecture-specialist","workflow-quality-specialist","knowledge-operations-specialist"] |
| category | retrieval |
| description | Actively retrieve context from the Dual-Storage Cognitive Memory system (Qdrant & SQLite) for seamless IDE integration. |
| knowledge | ["cognitive-memory-patterns.json"] |
| name | context-retrieval |
| related_skills | ["managing-memory"] |
| templates | ["none"] |
| tools | ["search_memory"] |
| type | skill |
| version | 1.0.0 |
| references | ["none"] |
| settings | {"auto_approve":true,"retry_limit":3,"timeout_seconds":60,"safe_to_parallelize":true,"orchestration_pattern":"standard"} |
Context Retrieval (Antigravity Architecture)
A cognitive agent must not operate in isolation. You have access to the search_memory tool provided by the RAG MCP Server. Use this skill to actively recall experiences, procedural knowledge, tools, and entities.
When to Use
- You are starting a new implementation plan and need to check how similar tasks were resolved.
- You hit a stack-trace or bug and want to check if the error is resolved in
memory_semantic.
- You interact with Plane MCP tools and need to recall specific task UUIDs from
memory_entity.
- You are looking for an exact factory workflow to execute in
memory_procedural.
Process
- Decide which memory tier contains the data type you need:
- memory_procedural: Workflows, Skills, Scripts.
- memory_toolbox: Capabilities, MCP actions, script descriptions.
- memory_summary: Post-session logs and high-level conversation synopses.
- memory_semantic: Condensed rules, facts, architectural patterns.
- memory_entity: Proper nouns, system names, file paths, UUIDs.
- Use the
search_memory MCP tool to retrieve documents from that specific collection.
- Inject the retrieved insights into your immediate task logic.
Example: Error Resolution
call:mcp_server:search_memory(query="ModuleNotFoundError scripts.memory PYTHONPATH", collection="memory_semantic")
Example: Finding a Workflow
call:mcp_server:search_memory(query="alpha factor mining", collection="memory_procedural")
Prerequisites
- The Qdrant Docker memory backend must be running on
localhost:6333.
- The
qdrant-rag MCP server must be active (provides search_memory tool).
- Access to the Python scripting environment in
/scripts/memory.
Best Practices
- Do not guess paths: If you need a script, retrieve its location from
memory_procedural or memory_toolbox.
- Pre-fetch: Always fetch context before proposing architectural designs in
implementation_plan.md.
- Tag the Memory: In the future when triggering tool calls (like creating an issue), ensure the description explicitly mentions the entities so the background
sessionEnd hook writes it clearly to the cognitive index.