searching-messages_skill

Overview

This skill searches past conversation messages to recall context that has fallen out of the active context window. It helps you locate previous discussions, verify prior statements, and identify which agent handled a topic. Use it to recover exact phrasing, timelines, or message IDs for deeper context expansion.

How this skill works

The skill performs indexed searches across stored messages using hybrid, vector, or full-text modes. Results include message_id, message_type, content or reasoning, created_at, and agent_id so you can expand around hits or trace the responsible agent. You can filter by date range, agent, and result limit to narrow results.

When to use it

• You need to answer “do you remember when we discussed X?”
• You must pull context from an earlier conversation not in the current window
• You want to verify what was said previously about a topic
• You need to find which agent discussed or authored a message
• You want to recover exact message text or a message_id for context expansion

Best practices

• Start with a concise keyword needle, then expand before/after the found message_id for full context
• Use hybrid mode for general searches, vector mode for conceptual matches, and fts for exact phrases
• Apply date bounds when you roughly know when the discussion occurred to reduce noise
• When tracing ownership, use --all-agents and cross-reference agent_id with agent-finding tools
• Limit results to a manageable number (default 10) and then expand around the most relevant hit

Example use cases

• Find the exact wording of a previous design decision to include in a spec
• Locate the conversation where an agent proposed a refactor to confirm rationale
• Recover a message_id to load surrounding messages and reconstruct a thread
• Identify which agent handled authentication discussion by searching across agents
• Search for a vague topic using vector mode when you lack precise keywords