memory-search
/memory-search
Search agent memory effectively: recall prior decisions, facts, and context across conversations. Use when an agent needs to remember what was learned before.
- Author
- doloopz94 ↗
- Repository
- virtual-office-muse-ai ↗
- Licence
- MIT
- Last repository change
- 11.11.2026
Assessment and security
- Files in this package
- 4
- Scripts
- 0
- HTML comments
- 0
- URLs in SKILL.md
- 0
No signals detected. Repository changed on 11.11.2026, without references to retired models or APIs.
MIT, open: use and modify while retaining author attribution.
What it does and when to use it
Search agent memory effectively: recall prior decisions, facts, and context across conversations. Use when an agent needs to remember what was learned before.
Original author description
Search agent memory effectively: recall prior decisions, facts, and context across conversations. Use when an agent needs to remember what was learned before.
How it works
- Overview
- When to use
- Core concepts
- Practical workflow
- Common pitfalls
SKILL.md preview
---
name: memory-search
description: Search agent memory effectively: recall prior decisions, facts, and context across conversations. Use when an agent needs to remember what was learned before.
category: ai-maestro
---
# Memory Search
## Overview
Agents without memory start every conversation from zero. Memory systems store durable facts, decisions, preferences, and history — searchable when needed.
Effective memory search: store well (atomic claims, sourced, dated), retrieve with multiple query phrasings, and verify before acting on old memories.
The pattern: write memories as you learn, search before deciding, and update when facts change.
## When to use
- Agent needs prior context about the user or project
- Recalling past decisions and their reasoning
- Finding what was learned in earlier sessions
- Verifying a fact before acting on it
- Maintaining continuity across long-running work
## Core concepts
- **Atomic claims.**
Store one fact per memory entry: who said it, when, the exact claim. Atomic entries retrieve precisely; paragraphs retrieve vaguely.
- **Source and date.**
Every memory notes its origin (conversation date, document) and timestamp. Old memories need age-checking before use.
- **Multiple query phrasings.**
Search with 2-3 different wordings of the same question. Retrieval is fuzzy — varied phrasings catch what one misses.
- **Verify before acting.**
Retrieved memories are leads, not gospel. Check the source context for anything consequential before acting on it.
- **Update, don't just add.**
When facts change, update or supersede the old memory. Contradictory memories without resolution poison future retrieval.
- **Salience and decay.**
Not all memories matter equally. Boost frequently-used, correct memories; let stale, unused ones fade. Memory needs gardening.
- **Scoped retrieval.**
Search the right store: user preferences vs. project facts vs. session history. Scoped search beats global noise.
- **Provenance chains.**
For important claims, keep the chain: memory -> source conversation -> original evidence. Auditability matters for trust.
## Practical workflow
1. **Capture as you learn.**
Durable facts, decisions, preferences -> memory immediately, in atomic form with source and date. Don't batch 'later.'
2. **Search before deciding.**
Before acting on prior work, preferences, or history: search memory with 2-3 phrasings. Make it a habit, not an afterthought.
3. **Read the context.**
For top hits, pull the surrounding source context. A claim without context is easy to misapply.
4. **Act on verified memories.**
Use verified memories to inform decisions; note the memory as the basis so the reasoning is traceable.
5. **Update on change.**
New information contradicting a memory? Update the entry (or mark superseded) immediately. Stale memories are worse than none.
6. **Garden periodically.**
Review memories: still true? Still relevant? Merge duplicates, retire the dead. Monthly for active stores.
7. **Handle conflicts.**
Two memories disagree: check sources and dates, prefer newer + better-sourced, record the resolution.
8. **Respect privacy.**
Memories contain personal data. Scope access, never leak across users, and honor deletion requests completely.
## Common pitfalls
- **Never searching.**
Having memory but deciding from scratch each time. The search habit is the whole value proposition.
- **Single phrasing.**
One query, no hits, giving up. Retrieval needs 2-3 varied phrasings — different words catch different entries.
- **Acting on stale memories.**
Using a year-old preference as current fact. Check dates; verify anything consequential.
- **Memory hoarding.**
Storing everything verbatim. Uncurated memory becomes unsearchable noise — atomic, salient, gardened.
- **Contradictions unresolved.**
Old and new memories disagreeing silently. Resolve explicitly or retrieval becomes a coin flip.
- **No provenance.**
Facts without sources. When challenged ('where did you learn that?'), sourceless memories collapse.
- **Cross-user leakage.**
One user's memories surfacing for another. Scope strictly; this is a privacy failure, not a feature.
- **Treating memory as truth.**
Memory is a lead, not evidence. Verify against sources for anything that matters.
Files in this package (4)
- ATTRIBUTION.md 1 KB
- INSTALL-NOTES.md 1 KB
- SKILL.md 4 KB
- skill.yaml 1 KB
Versions
| Repository | Score | Licence |
|---|---|---|
| This page · doloopz94/virtual-office-muse-ai | 90 | MIT |
| rjmurillo/ai-agents | 100 | MIT |
| rjmurillo/ai-agents | 100 | MIT |
| S3YED/appie-kit | 100 | MIT |
| S3YED/appie-kit | 100 | MIT |
| davila7/claude-code-templates | 95 | MIT |
| 23blocks-OS/ai-maestro-plugins | 95 | MIT |
| ibrain-bvba/gutt-claude-code-plugin | 90 | MIT |
| aristoapp/membase | 90 | MIT |
Database snapshot: 13.11.2026 · package version d1c3985a4baa (SHA-256 digest of the entire package)
Downloads in the last 30 days: 0 · ranking
Report a problem with this skill: harmful content, incorrect descriptions or an author’s removal request.
Similar skills
- 90recall
Search memories from past Claude Code sessions in the CURRENT project. Use when the user's question could benefit from historical context, past decisions, debugging notes, architectural choices, or prior conversations in this repo. For cross-project recall, use the recall-global skill instead. - 84memory
Store and retrieve durable agent memory — remember decisions, preferences, and session outcomes with calibrated salience; recall prior context before acting or resuming a project. Use whenever you store a memory, search memory, retrieve prior context, or check what was learned in past sessions. - 95paxm
Use paxm as Codex's active and passive memory layer. Trigger for prior decisions, project context, user preferences, memory recall, memory writes, or paxm diagnostics. - 95pact-memory
Persistent memory for PACT agents. Save context, goals, lessons learned, decisions, and entities. Semantic search across sessions. Use when: saving session context, recalling past decisions, searching lessons. Triggers: memory, save memory, search memory, lessons learned, remember, recall - 100memory
Persistent long-term memory via memini. Use to remember durable facts, decisions, and preferences, and to recall relevant context before acting. Triggers when the user says "remember", "what do you know about", or when starting a task that may have prior context. - 100agent-memory
Give an agent persistent semantic memory with @three-ws/agent-memory. Use when code needs to remember facts across sessions, recall them by meaning, read the entity graph, or curate an agent's working context.