researchlisted

# Research — answer it from the source that owns it Research is the **outside-knowledge** flow. Its whole job is to turn "nobody here knows" into a cited fact somebody can decide on. > Prior art: Matt Pocock's `research` skill (MIT). This is a re-derivation that wires the > flow into the playbook's memory, Serena and handoff conventions. ## Know which tool owns the question Getting this wrong is the expensive mistake. It burns a background agent rediscovering something you already own. | The question is about | Use | Not | |---|---|---| | **our** code — a symbol, a caller, a signature | Serena (`find_symbol`, `find_referencing_symbols`) | research | | something we already concluded once | **query memory first** | research | | a third-party API, spec, RFC, vendor doc | **research** | guessing from training data | | a ticket's own requirements | the tracker | research | **Query memory before you dispatch.** The two-memory rule means a prior investigation — especially a *negative* result — is already banked and one query away. Re-researching it is the exact cost the memory pillar exists to avoid. ## Dispatch it to the background Hand the question to a background agent and keep working. Research is read-heavy and context-hungry. It belongs in a throwaway context, like the gatherer's sweep. Give the agent the **question**, not a topic — "does provider X's webhook retry on a 500, and with what backoff?" not "look into webhooks". ## What the agent does 1. **Go to the prima