SessionEnd

~/.claude/settings.json

references/setup.md

Implementation Note: Since the

agent

type hook of

SessionEnd

has a silent failure bug (anthropics/claude-code#40010), switch to the

command

hook to call

scripts/summarize.sh

, and the script asynchronously completes the summary via

claude -p

internally.

• Only generated once for the same session (

  once: true

  )
• The file name contains the first 8 characters of session_id, and it will be overwritten instead of creating a new one when triggered repeatedly
• Automatically skipped if the number of conversation rounds is less than 3

templates/summary.md.template

1. Core Goal — What problem this session intends to solve
2. Completed Content — What was actually done, where the outputs are located
3. Key Decisions — What judgments worth recording have been made
4. Pending Items — Tasks that need to be continued next time
5. Skills Used — Which skills were called

• Public role: Conversation recording and context persistence
• Working directory:

  <project_root>/.claude/summaries/

• Recommended collocations:
  • Use with

    wechat-pipeline

    to record the execution status of each production line
  • Use with

    ai-workflow-brand

    to record each positioning decision