writing-plans

Implementation planning for multi-subsystem features works by breaking specifications into independent, testable sub-projects. It locks file decomposition boundaries and produces bite-sized TDD steps with concrete commands, ensuring clear execution handoff for any engineer.

Creating a TDD execution plan requires producing per-task, per-step instructions that include failing tests, command runs, minimal implementations, passing test verification, and frequent git commits. This ensures multi-file feature work is verifiable and isolated.

Generated implementation plans are saved to docs/superpowers/plans by default, with an optional override for custom output locations. This persistence enables subsequent review and execution handoff to subagents or inline workflows.

Implementation planning applies to multi-step feature work spanning multiple files and potentially multiple subsystems. It decomposes requirements into independent, testable sub-project plans with explicit file paths and concrete commands.

To hand off an implementation plan, generate a decomposed task plan with required headers, explicit file paths, concrete commands, and expected outcomes. The saved plan enables an unfamiliar engineer to execute end-to-end via subagent or inline execution.