AgentSkillsCN

remember

搜索过去的对话历史,以回忆之前的讨论、决定和上下文。当需要参考之前会话中讨论或决定的内容时使用。

SKILL.md
--- frontmatter
name: remember
description: Search through past conversation histories to recall previous discussions, decisions, and context. Use when you need to reference what was discussed or decided in previous sessions.

Conversation History Search

Search through past conversation histories stored in ~/.claude/projects/*/*.jsonl to help recall previous discussions, decisions, code explanations, and context from earlier sessions.

Instructions

  1. Run the search command with a query string:

    bash
    uv run python .claude/skills/remember/remember.py "<query>"
    
  2. Optionally include agent conversations (sub-tasks):

    bash
    uv run python .claude/skills/remember/remember.py "<query>" --include-agents
    

Features

  • Searches both user and assistant messages
  • Finds discussions about specific topics, features, or files
  • Shows conversation context with timestamps
  • Groups results by session for easy navigation
  • Sorts by recency (most recent first)
  • Skips agent conversations by default (use --include-agents to include them)
  • Case-insensitive substring matching

Examples

Search for discussions about networking:

bash
uv run python .claude/skills/remember/remember.py "networking"

Find conversations about a specific file:

bash
uv run python .claude/skills/remember/remember.py "DESIGN.md"

Look for combat system discussions:

bash
uv run python .claude/skills/remember/remember.py "combat system"

Search including agent conversations:

bash
uv run python .claude/skills/remember/remember.py "ENet" --include-agents

When to Use This Skill

  • Need to recall why a design decision was made
  • Want to find previous discussions about a feature
  • Looking for context about how something was implemented
  • Trying to remember what was decided in a past conversation
  • Need to reference code explanations from earlier sessions

Notes

  • Searches conversation histories in ~/.claude/projects/*/
  • By default, excludes agent conversations (background tasks)
  • Results are truncated to 200 characters for readability
  • Shows file path and line number for reference
  • Groups results by session to see conversation flow
  • Displays up to 50 most recent matches by default