Memory Recall

Problem

A new chat often gives the agent only the current message, so it can miss user preferences, active projects, past decisions, and known pitfalls. Asking users to restate context breaks flow and adds overhead. Memory Recall addresses this by letting the agent retrieve existing context before answering in a new session or task.

How It Works

The skill defines a four-step startup protocol that should finish before replying:
- Read long-term memory: inspect the long-term memory file; skip it if absent.
- Read recent logs: load today’s and yesterday’s log files to capture recent work.
- Search conversation history: call conversation_search once with a self-contained query that restates the current task, required context, and keywords.
- Fuse context: internalize user identity, active projects, safety rules, and past decisions, then use them to guide the response.

It expects silent execution: the agent should not announce recall. Preferences, decisions, and safety constraints should appear naturally. High-priority items include identity, active projects, safety rules, and destructive-operation guidelines; lower-priority items include tool preferences, naming conventions, and personal background.

Boundaries

This is useful for assistants that need continuity across sessions, project-aware agents, or coding assistants that must remember technical decisions and workflows. Missing files are skipped, empty search results fall back to file memory, and system instructions override recalled memory. The skill focuses on reading and fusing context; it does not write memory and should not recite memory unless directly relevant.