| name | reflection-and-planning |
| description | Activate this skill when beginning complex tasks, proposing architectural changes, or modifying codebase files to ensure proper planning and predictive failure analysis. |
Reflection and Planning Pipelines
To ensure thoughtful decision-making and robust implementations, the AI agent MUST strictly adhere to the following Reflection and Planning constraints before and after task execution.
1. Mandatory Implementation Plans
Before making structural modifications or editing logic blocks larger than 10 lines across any files, you MUST outline your approach using a formalized implementation plan.
- Artifact Creation: You MUST explicitly output an implementation plan using an
implementation_plan.md artifact.
- Approval Gate: You MUST set
request_feedback=true on the artifact, stop calling tools, and END your turn to await user approval.
- Prohibited Execution: You are FORBIDDEN from executing any code modifications using your file edit tools until the user responds explicitly with "Proceed" or approves the implementation plan.
2. Predictive Failure Analysis
Anticipating system failures before they happen is critical to stability. Whenever you finalize or execute code modifications, you MUST practice predictive failure analysis.
- Required Output Section: You MUST append a specific markdown section to your reasoning or visible output titled "Failure Analysis".
- Analysis Content: This section MUST detail exactly two explicit failure points (e.g., specific edge cases, unhandled promises, permission errors, or performance bottlenecks) related to the code you just wrote.
- Mitigation Strategy: You MUST explicitly describe exactly how your newly implemented code natively handles or mitigates these two predicted failure points.
Last Updated: 2026-07-22 | Last Reviewed: 2026-07-22