recall

Relevant memory was already injected at the top of this turn by the Mubit recall hook. Read it before searching. Most of the time you do not need this skill.

When you do:

1. Issue one broad mubit_recall call. Not three narrow ones.
2. Read the evidence. If it answers the question, stop.
3. Only if the first call returned nothing usable, issue one reformulated call.

Never fan out into parallel searches across sub-topics. Mubit retrieval is hybrid (semantic + lexical + recency + graph); one well-formed query beats four keyword slices at a quarter of the latency. Two calls is the ceiling.

Cite what you use by its reference_id, and call mubit_outcome with those entry_ids when recalled memory turns out to be right or wrong. That feedback is what makes the next recall better.

Writing the query

Query with the question, not with keywords. "Why does the drain hook retry twice on a 5xx" retrieves better than "drain retry 5xx", because the semantic half of the hybrid index has something to match on and the lexical half still catches the identifiers. Include the identifiers you already know — file names, symbol names, error strings — inside the sentence rather than instead of it.

A reformulation is a different concept, not a synonym. If "auth failed on ingest" returned nothing, "401 from the control plane" is a reformulation; "authentication failure ingest" is the same query with the words shuffled and will return the same nothing.

When two calls return nothing

Say so and move on. Empty is a real answer: it means nothing about this was ever captured, and a third query will not invent it. If the answer you then work out by hand is worth keeping, save it with /mubit-memory:remember so the next session's recall is not empty too.

Deep searches belong to the subagent

If the question genuinely needs several angles — "what do we know about X" across an unfamiliar area — invoke @mubit-memory:mubit-recall instead. It runs the same searches in an isolated context and returns a synthesis, so the raw evidence never lands here.