| name | remember |
| description | Store a memory for future retrieval. Use when the user says "remember this", wants to save context, preferences, decisions, or any information that should persist across conversations. |
| argument-hint | [memory text or description of what to remember] |
| allowed-tools | mcp__agent-memory__memory_store, mcp__agent-memory__memory_search, mcp__agent-memory__memory_working_set |
The user wants to store a memory. The input is:
$ARGUMENTS
Your job:
- Parse the input to determine what should be stored
- Apply the memory quality gate:
- Store only information that will help a future agent work faster: reusable patterns, operational procedures, user preferences, non-obvious constraints, failure causes, or "how to / why" guidance.
- Use
memory_working_set instead when the content is transient active-project handoff rather than durable knowledge.
- Write for a cold agent who has not seen this session: explain what to do next, which tool or system to use, and why that path is correct.
- Do not store facts recoverable from git history, repository inspection, CI/release systems, or agent-tools tasks/comms.
- Do not store routine deployment status, version numbers, release events, commit SHAs, branch state, "CI passed", "tag was pushed", or "deployed version X" memories.
- Store deployment/version facts only when they explain a failure mode or encode a reusable procedure that prevents future mistakes.
- Do not store "updated pattern X" notes. If a user refers to "patterns", they likely mean gateway-backed
agent-tools patterns; useful memories should describe the agent-tools patterns --help / get / update / check workflow and why it matters.
- Short locator memories for canonical gateway patterns are allowed when they help a cold agent quickly find and reuse non-obvious guidance. Keep them to the locator, reuse instruction, and why the pattern matters; do not record "created/updated pattern X" as an audit event.
- Choose the appropriate memory_type:
- "user" — facts about the user (role, preferences, expertise)
- "feedback" — guidance on how to approach work (corrections, confirmations)
- "project" — ongoing work context (decisions, deadlines, constraints)
- "reference" — pointers to external resources (URLs, systems, dashboards)
- Extract relevant tags (comma-separated, lowercase, specific)
- Detect the project context if obvious from the conversation or working directory
- Call the
memory_store MCP tool with:
content: the memory text, written clearly for future retrieval
tags: extracted tags
project: project identifier if known
memory_type: one of the four types above
agent: the agent identifier if this memory is agent-specific
- Confirm storage to the user with the memory ID
If the input is vague, ask a clarifying question before storing. Memories should be specific enough to be useful when retrieved later.
If the memory is a correction or preference about how you should behave, make sure to frame it as actionable guidance in the content field.