Please let us add custom instructions/rules for ToDo lists.
Itâs also important they can be workspace level for continuing in new chats.
I was using progress tracking and session management rules to chunk tasks by tool calls and write documentation for hand-off to a fresh chat when the context window was getting full.
Example of one of my rules
Summary
Automatic Progress Tracking & Documentation
This rule ensures consistent progress tracking and automatic updates to project documentation during development work.
Automatic NEXT_STEPS.md (or other specified file) Updates
ALWAYS update NEXT_STEPS.md (or a different file if specified) when completing work
Keep completed items in NEXT_STEPS.md if user testing/feedback is still needed
Progress Status Classifications:
# In NEXT_STEPS.md:
- â **COMPLETED** - Feature implemented, needs user testing
- đ§ **IN PROGRESS** - Currently being worked on
- â¸ď¸ **PAUSED** - Partially complete, waiting for user input
- đ **PLANNED** - Next item to work on
# Move to HISTORY.md only after:
- User confirms testing is complete
- User provides feedback that feature works as expected
- No further iteration required
Update Format:
â **Feature Name** - COMPLETED (Needs Testing)
- **Implementation**: Brief summary of what was built
- **Status**: Ready for user testing and feedback
- **Next Steps**: User to test functionality and provide feedback
Context Window Management
Tool Call Counter:
Track tool calls throughout conversation
At 20-22 tool calls, proactively update progress documentation
At 24-25 tool calls, provide comprehensive status summary
Progress Checkpoints:
After every 20+ tool calls, update NEXT_STEPS.md with:
- Current implementation stage
- What has been completed
- What is currently in progress
- Next immediate steps
- Estimated remaining effort
đ§ **Current Stage**: Stage 2 - Frontend UI Components
- â Component structure created
- â Basic functionality implemented
- đ Currently working on: State management
- đ Next: Event handlers and validation
Progress Documentation Requirements
Real-time Updates:
Update progress in NEXT_STEPS.md as work proceeds
Use specific status indicators ()
Include current work context for easy resumption
End-of-Session Summary:
## đ Current Session Progress
### Latest Updates:
- [x] **Feature A**: Backend implementation complete â
- [x] **Feature B**: Frontend UI 50% complete đ§
- [ ] **Feature C**: Planned for next session đ
### Current Status:
- **Active Work**: Frontend component integration
- **Tool Calls Used**: 22/25 this session
- **Next Steps**: Complete state management, then user testing
- **Estimated Remaining**: 1-2 requests
Chat Handoff Preparation
Context Preservation:
Always update NEXT_STEPS.md before hitting tool call limits
Include specific implementation details and current progress
Document any discovered issues or approach changes
Provide clear next steps for chat continuation
Handoff Checklist:
Before reaching 25 tool calls:
1. â Update NEXT_STEPS.md with current progress
2. â Document any code changes made
3. â Note current implementation approach
4. â List immediate next steps
5. â Estimate remaining effort
User Testing Integration
Hold in NEXT_STEPS.md Until:
User confirms feature works as expected
User provides feedback on any needed changes
All iterations and refinements are complete
Testing Status Updates:
â **Feature Implementation** - COMPLETED (Awaiting User Testing)
- **Last Updated**: [timestamp]
- **Status**: Ready for user testing
- **Testing Notes**: User should test X, Y, Z functionality
- **Known Issues**: None currently identified
Best Practices
Proactive Documentation: Update progress before being asked
Transparent Progress: Always show current stage and remaining work
Context Continuity: Ensure any new chat can pick up exactly where left off
Efficient Handoffs: Minimize re-explanation needed between chats
Clear Status: Use consistent status indicators throughout documentation
Remember: The goal is seamless progress tracking that prevents lost context and reduces the need for manual documentation requests. Better to complete 80% of a feature with perfect documentation than 100% with poor handoff context.
that setting does not work in 1.2 due to the new VSCode base version used in 1.2.
I understand not supporting issues after the marketplace is changed.
I do not understand not supporting changing marketplaces.
Use 3.7 Thinking if youâre on the pro plan with new rate usage etc. Once you use custom rules and commands etc. and make it use flow charts in ASCII it follows perfectly and correctly. Zero errors.
My use case: Almost always Sonnet, sometimes Opus for harder tasks.
Switching to large context probably unnecessary:
Edit: Isaac - Removing your comment on max switch as itâs misleading. Max can still be used on all plans, you donât need usage-based pricing to use it.
Oh dam n, then Iâve misunderstood the pricing.. I was confidently feeling smug that I got it, whilst everyone else questioned it.
I understood it to be:
Pay $20 a month, get âgoodâ access, which will dwindle in efficiency as you use up the rates. Pay extra usage-based fees for MAX, which includes BG agents.
Increase the amount of âgoodâ access x3 or x20 with the Pro+ or Ultra plans.
Iâve presumed that BG agents pricing considerations were synonymous with MAX:
after 2+ hours of working with the app yesterday not a single memory was generated dont know if its broken or the agent did not find important enough information to update
You can try a network diagnostic test from the cursor settings â Network tab
Might be a proxy or VPN, disabling HTTP/2 can sometimes bypass these errors if those are the root cause. Either that or some antivirus/firwall might be blocking outbound requests.
If youâre able to share a requestID I can see if it hit our backend or not (to confirm a local network/client issue)
"Max Mode is included in all plans. Turning it on will pay down your agent limits slightly faster, since the API cost of your requests will go up a bit. "
Usage-based pricing can still also be enabled on pro plans, it will be used when rate limits are hit, and weâre working on improving messaging around when this usage will be consumed in-app. BG agents IIRC require enabling usage-based pricing with $10, and currently use MAX models for their usage, but we may remove this requirement in the future for needing usage-based pricing enabled to use it.
It seems the language server issues was because Open VSX Marketplace did not work in 1.2 update at all (probably due to VSCode version update broke something with marketplaces), so I had to default to Cursor Marketplace by deleting this part from User Settings (JSON):
You have to manually tell the AI/LLM etc. to use the memory_tool to make a memory out of contextual awareness. It will also update the memories based on the conversation as long as you wait several minutes etc.
ALWAYS update NEXT_STEPS.md (or a different file if specified) when completing work
Progress Status Classifications:
Update Format:
Tool Call Counter:
Effort Estimation in âRequestsâ:
Stage-Based Implementation:
Stage Documentation:
Real-time Updates:
Efficient Handoffs: Minimize re-explanation needed between chats
