- Quick Reference
- Introduction
- General Guidelines
- Workflow Progression Prompts
- Project Status Prompts
- Git and Version Control Prompts
- Team Coordination Prompts
- Analysis and Review Prompts
- Context Management Prompts
- Creative Content Prompts
- Best Practices
- Version History
Most frequently used prompts for common tasks:
## Standard Progression
Reason and proceed with the most logical action(s).
## Status-Aware Check
Briefly analyze the project status, reason, and proceed with the most logical action(s).
## Commit Changes
Now commit all changes in an organized manner.
--- If `git diff` is required, use it for all changes, saving the output in a temp file to facilitate the reading.
--- Always use conventional commits standards.
## Team Note Review
Read all team notes you haven't read yet. Always start by checking @_read_receipts.txt. After reading each note, reason briefly, but don't take actions besides the notetaking protocol.
After that, reason about them, and if really necessary, take appropriate actions.
This guide provides standardized prompting patterns for effective interaction with AI assistants throughout the InvoiceForge project. Following these patterns ensures consistent outputs, maintains context across interactions, and helps the AI deliver more relevant, actionable responses tailored to specific workflow stages and tasks.
- Be specific and concise in your instructions
- Structure complex prompts with clear sections
- Include necessary context when switching topics
- Use a consistent prompt style within each workflow category
Use these prompts to guide the AI through the project's TDD workflow and ITC process.
## Standard Progression
Reason and proceed with the most logical action(s).
## Status-Aware Progression
Briefly analyze the project status, reason, and proceed with the most logical action(s).
## Analysis Before Action
Briefly analyze our current state in the 5-Stage TDD Workflow, then proceed with the most logical action(s).
## ITC-Focused Analysis
Briefly analyze our current state in the 6-Step ITC, then proceed with the most logical action(s).
## Modified ITC Workflow
We modified application code.
Briefly analyze our current state in the 6-Step ITC, then proceed with the most logical action(s).
You can skip linting this time.
## Documentation Check
Check the conclusion of the ITC section in @development.md. Reason and proceed with the most logical action(s) accordingly.
- Standard Progression: For general progression when the context is clear and you want the AI to determine and take the next appropriate action.
- Status-Aware Progression: For periodic check-ins on overall project progress that should lead to appropriate next actions.
- Analysis Before Action: When you want explicit analysis of where you are in the TDD process before proceeding.
- ITC-Focused Analysis: Specifically for tracking progress within the Iterative Testing Cycle.
- Modified ITC Workflow: After code changes when you want to follow up with specific testing steps but skip certain parts of the cycle.
- Documentation Check: When you need to align actions with documented procedures.
Use these prompts to get a comprehensive view of the project status and plan next steps.
## Planning-Focused Review
Reason and plan actions.
## Agent Workflow Review
Now analyze the project status and the agent workflow. Then reason and proceed with the most logical action.
- Planning-Focused Review: When you need a structured plan before taking action.
- Agent Workflow Review: To assess how the multi-agent workflow is functioning and identify appropriate next steps.
Use these prompts for source control management and documentation updates.
## Commit Changes
Now commit all changes in an organized manner.
--- If `git diff` is required, use it for all changes, saving the output in a temp file to facilitate the reading.
--- Always use conventional commits standards.
## Changelog Updates
Analyze the information from the last 20 commits (with Git commands) and update CHANGELOG accordingly. Remember to keep a user-focused language and to filter irrelevant info.
## Documentation Updates
Now analyze the project status, then update development_roadmap.md (mandatorily), README (optionally), @implementation_plan (optionally), and any other necessary document accordingly.
## Git Commands Usage Note
(Never leave the terminal in interactive mode – Use suffix ` | cat` when needed. Use the fewest commands as possible.)
- Commit Changes: When you need to commit multiple changes with proper organization and documentation.
- Changelog Updates: When preparing releases or updating documentation to reflect recent changes.
- Documentation Updates: For periodic documentation refresh to keep project artifacts in sync with implementation.
- Git Commands Usage Note: Add to any prompt containing Git commands.
Use these prompts for communication and coordination between agents.
## Agent Boundary Check
Do you think this work is venturing too far into the domains of another agent? If so, pass the baton to this agent with a message. If not, keep up the work.
## Team Note Review
Read all team notes you haven't read yet. Always start by checking @_read_receipts.txt. After reading each note, reason briefly, but don't take actions besides the notetaking protocol.
After that, reason about them, and if really necessary, take appropriate actions.
## Message Response
Read message @msg, reason, and take appropriate action(s) if necessary.
## Restricted Communication
Read message @msg, reason, and, if necessary, take appropriate actions. Communicate the team (via message(s) and/or a single team note) only if really necessary.
## Information-Only Messages
Read message @msg and reason. Do not respond to the sender.
- Agent Boundary Check: When you suspect that a task might be better handled by another specialized agent.
- Team Note Review: To catch up on team communications and determine if action is needed.
- Message Response: When responding to specific messages that may require action.
- Restricted Communication: When you want to limit communication to only essential updates.
- Information-Only Messages: When you need the AI to understand information without generating a response.
Use these prompts for code and documentation review.
## Coherence Analysis
Analyze the whole [codebase/documentation] for coherence.
## Analysis with Action Planning
Save your analysis for future reference, then review your proposed actions and apply the ones with high gain and low effort.
## Agent Definition Reviews
Read the git diff of your agent's definition file (XXXXX.cursorrules) on the last commit (COMMIT_HASH). Use a single diff command, saving the result in a temp file. Reason about and, if strictly necessary, take actions.
- Coherence Analysis: For comprehensive review of project artifacts to ensure alignment and consistency.
- Analysis with Action Planning: When you need thorough analysis followed by prioritized actions.
- Agent Definition Reviews: When reviewing changes to agent definitions and determining if updates are needed.
Use these prompts when managing the AI's working context.
## Context Reset
Your context is getting too heavy and interacting with you is becoming too slow. Write a KT for a next agent with the same role to continue the work.
## Agent Handoff
[@XXX is your system message. @agent_workflow.md defines the agent workflow you belong to. @development_roadmap.md is the development roadmap with current status.]
The previous Agent X just sent you this report: @msgXX. They're passing the baton to you. Agent Y also sent you a message: @msgXX.
To help you with the outlined tasks and goals, I suggest also analyzing the most relevant docs under docs/ and docs/analyses, exploring code directories, and analyzing the current application codebase.
## Environment Setup
Run `source ~/.zshrc` once before using conda command.
## Terminal Troubleshooting
There's an issue with Cursor's terminal in which it got stuck in an old output. We may need to restart the tool (Cursor's terminal). After the restart, use `source ~/.zshrc`.
- Context Reset: When the conversation history becomes too long and slows down response time.
- Agent Handoff: When transitioning work from one agent instance to another.
- Environment Setup: When environment configuration is needed before executing commands.
- Terminal Troubleshooting: When addressing known terminal issues.
Use these prompts for generating creative assets and content.
## Image Generation
Write a prompt directed to a text-to-image AI asking to generate a suitable cover image for README.
- Image Generation: When you need to create visual assets for project documentation.
- Maintain Context: Reference specific files with @ symbol (e.g., @development.md)
- Be Explicit: Specify exactly what you want the AI to analyze or focus on
- Structure Complex Tasks: Break multi-step processes into clear sections
- Limit Scope When Needed: Use qualifiers like "strictly necessary" or "if needed" to prevent overaction
- Reference Documentation: Point to specific documentation when alignment is crucial
| Version | Date | Changes |
|---|---|---|
| 1.0 | 2025-03-12 | Initial version of prompting guide with blockquote format |
| 1.1 | 2025-03-12 | Enhanced with code blocks, TOC, and quick reference |
| 1.2 | 2025-03-12 | Consolidated code blocks for a more compact format |
| 1.3 | 2025-03-12 | Improved readability with consistent vertical spacing |
This guide should evolve as the project progresses and new effective prompting patterns emerge. Update it regularly to ensure it remains a valuable resource for the entire team.