| name | memory-recall |
| description | Search past session summaries and topic files for context relevant to current work. Use when resuming work on a previously built agent, continuing a prior session, or when the user references past decisions. |
Memory Recall
Search the project's session memory for context relevant to the current task.
When to Use
- The user references a previous session or agent they built before
- You're about to start work on an agent that may have prior history
- The user asks "what did we decide about X" or "remind me about Y"
- MEMORY.md mentions an active thread related to the current task
How to Search
-
Start with the index. Read .claude/memory/MEMORY.md for the overview of past work, active threads, and agent history.
-
Search session summaries. Use Grep to search .claude/memory/sessions/ for relevant keywords:
- Agent names (e.g.,
transport-agent)
- Phase names (e.g.,
build, validate)
- Specific topics (e.g.,
deployment target, hook pattern)
- Tags in YAML frontmatter
-
Read matching files. Open the 2-3 most relevant session summary files. Focus on:
- Decisions made
- Open threads / follow-up items
- Resume prompts
-
Check topic files. If the query relates to a recurring pattern, check .claude/memory/topics/ for consolidated learnings.
What to Extract
From the matching memories, extract only:
- Decisions that affect current work
- Constraints or conventions to follow
- Unresolved issues that need attention
- Artifact paths the user may need
- Resume context (where to pick up)
What NOT to Do
- Do not read every session file — search first, read selectively
- Do not inject raw transcript content
- Do not surface irrelevant history just because it exists
- Do not read more than 3-4 session files per recall
