writing-plans

Invoke the writing-plans skill with your spec or requirements ready. It produces a plan document with a header (goal, architecture, tech stack, global constraints), then tasks with exact file paths, interface contracts, and TDD steps saved to docs/superpowers/plans/.

Each step should be one action taking 2-5 minutes, such as writing a failing test, running it, implementing minimal code, verifying the pass, and committing. Tasks fold setup and docs into the deliverable that needs them and end with an independently testable result.

Split when the spec covers multiple independent subsystems. Each plan should produce working, testable software on its own, so one plan per subsystem keeps scope reviewable and executable.

Subagent-driven execution dispatches a fresh subagent per task with two-stage review between tasks, while inline execution runs tasks in the current session in batches with checkpoints. The plan handoff step offers both options after saving.

Plans fail when they contain placeholders like TODO or vague steps such as add error handling, reference undefined types, or have inconsistent names across tasks. The skill's no-placeholder rules and self-review checklist for spec coverage and type consistency address these failure modes.