| name | reflect-on-conversation |
| description | Analyze conversation history to provide structured retrospective on user prompting, agent performance, system gaps, efficiency, and technical decisions. Use when wanting insights to improve collaboration, optimize workflows, or understand what worked/didn't work. |
Conversation Reflection & Retrospective
Overview
Analyze the entire conversation history up to this point and provide a comprehensive, structured retrospective. The goal is to improve collaboration between the User and the Agent, optimize system configuration, and refine the user's prompting strategy.
Analysis Dimensions
When conducting the reflection, thoroughly examine each of these dimensions:
1. User Prompting & Guidance
Examine:
- Clarity & Intent: Were the user's initial requests and subsequent instructions clear? Could they have been phrased more effectively?
- Context Provision: Did the user provide enough context (files, background, constraints) upfront? What context was missing that caused delays?
- Steering: How effectively did the user guide the agent? What steering strategies worked best?
- Actionable Feedback: Suggest specific rephrasings or prompt patterns for future tasks.
- Clarifying Questions: Were the agent's questions to the user effective and necessary?
2. Agent Performance & Technical Execution
Examine:
- Reasoning Quality: Did the agent break down complex problems effectively? Where did reasoning falter?
- Tool Usage: Did the agent use the right tools? (e.g.,
codebase_search vs grep, parallel vs sequential calls)
- Proactivity: Was the agent proactive in identifying issues/improvements?
- Rule Adherence: Did the agent follow
.cursor/rules or CLAUDE.md? Are there conflicts?
- File Navigation: Was file discovery efficient?
- Code Quality: Did the agent adhere to best practices and patterns?
- Testing & Validation: Were changes verified? Were tests run or created?
3. System, Documentation & Environment
Examine:
- Documentation Gaps: What knowledge was missing? Suggest new
.mdc files, documentation, or updates.
- Tooling Needs: What manual tasks could be automated (scripts, CLI tools, MCP servers, skills)?
- Context Window: Was relevant code easily accessible?
- Knowledge Base: What tribal knowledge was uncovered that should be documented?
4. Process & Efficiency
Examine:
- Efficiency: Where were time/tokens wasted? Could the solution have been reached in fewer steps?
- Rabbit Holes: Did the conversation diverge? Identify the specific message index where drift began.
- Planning vs Execution: Was the plan adequate? Should planning have been more thorough?
- Scope Management: Was the task scope maintained or did it creep?
5. Technical Retrospective & Alternatives
Examine:
- Alternative Paths: Knowing what we know now, was there a better library, pattern, or architectural approach?
- Decision Points: Were there key decisions where alternatives should have been discussed?
- Automation: Are there specific patterns in the work that suggest a need for automation (skill, script, MCP server)?
Output Format
Organize your reflection starting with these three prioritized sections, followed by detailed analysis:
🛠️ Top 3: New Skills, Tools & Tooling
The 3 most impactful ideas for new scripts, tools, MCP servers, or Cursor skills to build.
- [Tool/Skill Name]: [Specific description of what it would do and why it would help]
- [Tool/Skill Name]: [Specific description of what it would do and why it would help]
- [Tool/Skill Name]: [Specific description of what it would do and why it would help]
🚀 Top 3: Prompting & Context
The 3 most critical pieces of advice for the user on how to ask better questions or provide better context.
- [Prompting Pattern]: [Specific advice with before/after examples if applicable]
- [Context Strategy]: [Specific advice with before/after examples if applicable]
- [Steering Technique]: [Specific advice with before/after examples if applicable]
✨ Top 3: Other Suggestions
The 3 most significant other suggestions (process, documentation, agent behavior, etc.).
- [Suggestion Category]: [Specific recommendation and rationale]
- [Suggestion Category]: [Specific recommendation and rationale]
- [Suggestion Category]: [Specific recommendation and rationale]
🤖 Agent Course Correction
Self-correction on what the agent should have done differently (reasoning, tooling, proactivity).
- Identify specific moments where the agent:
- Used the wrong tool or approach
- Failed to ask clarifying questions when needed
- Made incorrect assumptions
- Could have been more proactive
- Missed obvious solutions or patterns
📚 Documentation & Knowledge
Specific recommendations for new or updated documentation. Identify context gaps.
- Suggest new documentation files or updates to existing ones
- Identify tribal knowledge that should be captured
- Recommend additions to
.cursor/rules or CLAUDE.md
- Note any external documentation (libraries, APIs) that would have helped
🔄 Workflow & Process
Observations on efficiency, planning, and error recovery. Highlight any "rabbit holes" or divergence points.
- Identify where the conversation was most/least efficient
- Note any planning gaps that caused rework
- Highlight divergence points and how to avoid them
- Suggest process improvements for similar tasks
💡 Technical Retrospective
Alternative technical approaches or patterns that could have made the task easier.
- Were there better libraries or tools available?
- Could a different architectural pattern have simplified things?
- Are there emerging best practices that should be adopted?
- What technical debt was created and how to address it?
✅ Next-Time Checklist
A short bulleted checklist tailored to this specific type of task for future reference.
For [describe task type] tasks:
Guidelines
Be Specific: Avoid generic advice. Reference specific message numbers, file paths, or command sequences.
Be Honest: If the agent made mistakes, identify them clearly. If the user's prompts were unclear, explain why.
Be Actionable: Every suggestion should be concrete and implementable.
Focus on Patterns: Look for recurring issues or successful strategies that can inform future work.
Consider Context: Account for the user's expertise level, project complexity, and time constraints.
