Search Memory

by nowledge-co1f4d7eb61466dbfc93ce8d7a0a806c4d4a196dd0No licenseListed Oct 9, 2026Updated Oct 9, 2026

Route recall across memories and threads when past work would improve the response.

Instructions onlyAI & Agents

AI-generated overview

Routes recall across stored memories and past threads when prior work would improve an answer.

What it does: This skill defines when and how an agent should search prior knowledge before responding. It lists strong and contextual triggers for recall, such as references to earlier fixes, decisions, or named features, and specifies a retrieval order: memory search first, then thread search, then progressive message fetching. It advises using the smallest retrieval surface that answers the question and searching early during continuation-heavy engineering work.
When to use it: Use it when a task resumes earlier work, references a prior decision or fix, resembles a previously solved debugging pattern, or asks about rationale, preferences, or established procedures. It also applies to reviews, regressions, releases, docs-alignment, and connector-behavior questions where past context matters.
Requirements: It relies on memory and thread search tools, including memory search, thread search, and thread message fetching, and assumes access to stored memories and conversation history. It ships no scripts; it is instructions only.

Use Nowledge Mem proactively when prior knowledge would materially improve the answer.

Search when:

the user references previous work, a prior fix, or an earlier decision
the task resumes a named feature, bug, refactor, incident, or subsystem
the task is a review, regression, release, docs-alignment, or connector-behavior question
a debugging pattern resembles something solved earlier
the user asks for rationale, preferences, procedures, or "how we usually do this"
the user uses implicit recall language: "that approach", "like before"

Contextual signals — consider searching when:

Start with memory_search for durable knowledge.
Use thread_search for prior discussions, previous sessions, or exact conversation history.
If a memory result includes source_thread_id, or thread search finds the likely conversation, use thread_fetch_messages progressively.
Prefer the smallest retrieval surface that answers the question.

Avoid over-reading long conversations when one page of messages is enough.

For continuation-heavy engineering work, search near the start of the task rather than waiting for an explicit recall request.

Source and attribution

License: No license

Content belongs to its original authors. SourceWeft indexes it from a public repository.