MN755 Mission Control RAG Retrieval Design
Purpose
Design retrieval systems that support agent work without turning context into a random pile of embeddings.
The Codex chat agent is not the Mission Control Manager. It is the bridge between the user and the Mission Control Manager.
Use when
- The user wants semantic search, RAG, retrievers, or context packs.
- Large docs or codebases need selective context.
- Retrieval quality needs evaluation.
Workflow
- Ask Mission Control to define corpus, chunking, metadata, ranking, and freshness rules.
- Choose local-first or API-backed retrieval based on policy.
- Define context-pack boundaries and citation expectations.
- Plan retrieval evaluation with representative questions.
Mission Control calls
Tools:
mission_control_start_taskmission_control_get_status
Resources:
mission-control://projects/{project_id}/codebase-mapmission-control://projects/{project_id}/agent-contractsmission-control://projects/{project_id}/validation-summary
User-facing output
- Summarize corpus scope, retrieval design, freshness, privacy posture, eval plan, and limitations.
Approval behavior
Ask before sending private content to external embedding or vector services.
Never do
- Do not imply semantic search is accurate without evals.
- Do not leak secrets into indexes.
- Do not use stale indexes without saying so.
Failure and fallback
If no retriever exists, propose a simple lexical baseline before vector search.
Example invocation
Use Mission Control to design a local-first RAG path for this docs folder.