Post-Mortem Analysis

Analyze chat histories to identify successes, failures, and improvement opportunities. Generate actionable recommendations for updating project configuration files (.cursorrules, skills, etc.) to prevent similar issues.

When to Use

• User asks for a "post-mortem" or "retrospective"
• User wants to analyze what went well or wrong in a session
• User wants to improve agent behavior based on past interactions
• User provides a chat export file or URL for analysis

Input Sources

The analysis can work with:

1. Current session: Analyze the ongoing conversation (default if no file provided)
2. File path: Chat export in JSON, markdown, or text format
3. URL: Link to a shared chat export

Process

Phase 1: Analysis

1. Load and analyze chat:

   • If no file specified: Analyze the current conversation history
   • If file path provided: Read the chat export file
   • If URL provided: Fetch the chat export using WebFetch
   • Parse the conversation flow and identify key interactions
   • Extract tool usage patterns and decision points
   • Note any error messages, confusion, or repeated attempts

2. Categorize interactions:

   • Successful patterns: What worked well and why
   • Failed patterns: What went wrong and root causes
   • Missed opportunities: Where the agent could have been more effective
   • User friction points: Where the user had to provide additional guidance

Phase 2: Assessment

1. Skill usage review:

   • Identify skills that were (or should have been) applied
   • Locate and read skill files in .cursor/skills/ or ~/.cursor/skills/
   • Assess whether skill instructions were clear and complete

2. Identify root causes:

   • Missing context in project rules or skills
   • Unclear or ambiguous instructions
   • Missing skills or workflow guidance
   • Tool selection issues
   • Architecture understanding gaps

3. Pattern analysis:

   • Recurring mistakes or confusion
   • Successful strategies that should be reinforced
   • Dependencies that weren't clear
   • Workflow steps that were skipped or misunderstood

Phase 3: Recommendations

1. Project rules improvements (.cursorrules):

   • Missing architectural context to add
   • Workflow patterns to emphasize
   • Common pitfalls to warn about
   • Examples that would clarify usage

2. Skill enhancements:

   • New skills that should be created
   • Existing skills that need updates
   • Additional constraints needed
   • Process clarifications
   • Tool usage guidelines

Phase 4: Implementation

1. Present findings:

   • Summary of what went well
   • Key issues identified
   • Specific recommendations with rationale
   • Proposed changes to configuration files

2. User confirmation:

   • STOP: Present findings and ask "Review these recommendations. Proceed with updating files? (y/n)"
   • Allow user to modify or reject specific recommendations

3. Apply improvements:

   • Update .cursorrules with approved changes
   • Modify or create skills as needed
   • Document changes made

Example Interactions

Analyze current session:

"Do a post-mortem on this chat"

1. Review current conversation history
2. Identify issues or missed opportunities
3. Present recommendations
4. Get confirmation before updating files

Analyze external chat:

"Post-mortem this chat export ~/Downloads/session.json - focus on the build process issues"

1. Read the chat export file
2. Focus analysis on build-related interactions
3. Identify missing documentation or skills
4. Present recommendations
5. Get confirmation before updating files

Output Checklist

• Analysis report: What went well vs. what went wrong
• Root cause analysis: Why issues occurred
• Actionable recommendations: Specific file changes to prevent recurrence
• Updated configuration: Improved rules and skills (after user approval)