Perform an end-of-session wrap-up. Review the session and provide actionable suggestions.
Steps
1. Pattern Analysis
Review the conversation history for:
- •Commands or operations repeated 3+ times (candidates for skills/aliases)
- •Manual processes that could be automated
- •Questions asked that indicate missing project context
Be medium-aggressive with suggestions: if something was done multiple times and would benefit from automation, suggest it. Don't suggest skills for one-off tasks.
2. Workflow Suggestions
Based on patterns observed, suggest (in text, not auto-applied):
Potential new skills:
- •If a multi-step process was repeated, propose a skill with the steps
- •Format: skill name, description, what it would do
claude.md additions:
- •Project conventions discovered during the session
- •Preferences the user expressed
- •Context that would help future sessions
- •Directly edit CLAUDE.md to add these (read it first, then append to appropriate section)
Other improvements:
- •Bash aliases or shell functions
- •Permission additions for settings.local.json
- •Subagent configurations for common tasks
3. Prompting Feedback
Review the user's prompts throughout the session and provide constructive feedback:
Clarity & Specificity:
- •Were requests clear enough to act on immediately?
- •Did vague requests cause back-and-forth clarification?
- •Could the user have provided more context upfront?
Efficiency:
- •Did the user break large tasks into reasonable chunks?
- •Were there unnecessary intermediate prompts that could be combined?
- •Did the user leverage Claude's capabilities effectively (e.g., parallel tasks, batch requests)?
Best Practices:
- •Note any particularly effective prompts as examples
- •Suggest specific rephrasing for prompts that could be improved
- •Highlight when more/less detail would have helped
Be honest but constructive. The goal is to help the user become more effective at working with Claude.
4. Pickup Note
Write a brief 2-3 sentence note summarizing:
- •Where work left off
- •Any blocking issues or open questions
- •Suggested next steps
Output Format
## Workflow Suggestions ### Potential Skills [Skill suggestions or "None identified"] ### claude.md Additions [List what was added to CLAUDE.md, or "None identified"] ### Other Improvements [Aliases, permissions, etc. or "None identified"] ## Prompting Feedback [Honest, constructive feedback on the user's prompts with specific examples and suggestions] ## Pickup Note [Brief note for resuming next session]
Keep the output concise and actionable. Only suggest improvements that would provide real value based on observed patterns.