AgentSkillsCN

summarize-conversation

由于内存限制，总结对话内容，并在简短日志中继续推进后续工作。

SKILL.md
--- frontmatter
name: summarize-conversation
description: summarize conversation due to memory it and work to continue while short log.
user-invocable: true
allowed-tools: Read, Grep, Glob

Skill: Conversation Summarizer (Context Memory)

Purpose

When requested, summarize the current conversation into a compact, reusable context that can be pasted at the beginning of a new conversation and understood by Claude without the original chat history.

This summary is intended to:

  • •Preserve important decisions and assumptions
  • •Reduce token usage
  • •Allow the current conversation to be safely cleared
  • •avoid to long wait for Claude to read past long conversation

When to Activate

Activate this skill ONLY when the user explicitly requests one of the following:

  • •"summarize this conversation"
  • •"create a context memory"
  • •"prepare a summary for the next conversation"
  • •"save this conversation in a reusable form"

Do NOT activate automatically.

Output directory and file name

${PROJECT_ROOT}/.claude/last-conversation.md Overwrite file everytime.

Output Format

Output MUST follow this exact structure:

Context Memory

Project / Topic

  • •Brief description of the project or discussion topic

Goals

  • •What the user is trying to achieve

Key Decisions

  • •Explicit decisions already made
  • •Chosen approaches or rejected alternatives

Constraints & Rules

  • •Technical, design, or conceptual constraints
  • •Non-negotiable rules or preferences

Terminology

  • •Definitions of important terms or abbreviations (if any)

Open Questions

  • •Unresolved issues or decisions not yet finalized

Style Rules

  • •Use concise bullet points
  • •No conversational tone
  • •No speculation or new ideas
  • •Do NOT repeat the full conversation
  • •Do NOT include timestamps or speaker labels
  • •Prefer factual, stable information only

Length Limit

  • •Target: 300–600 characters
  • •Absolute maximum: 800 characters

Assumptions

  • •Assume the summary will be pasted verbatim into a new conversation
  • •Assume no other context is available
  • •Optimize for Claude’s comprehension, not human storytelling

Forbidden Content

  • •Emojis
  • •Apologies
  • •Meta commentary (e.g., "In this conversation we talked about...")
  • •Suggestions beyond what was already discussed