writing-plans

Start with a plan header stating goal, architecture, and tech stack, then map the file structure before defining tasks. Break work into bite-sized TDD steps: write the failing test, verify it fails, implement minimally, verify it passes, and commit.

Each step should be one action taking 2-5 minutes, with exact file paths, complete code blocks, exact commands, and expected outputs. Tasks must be self-contained so an engineer reading them out of order can still execute them.

Split when the spec covers multiple independent subsystems. Each plan should produce working, testable software on its own, ideally one plan per subsystem identified during the brainstorming phase.

Plans fail when they contain placeholders like TODO or vague steps such as add error handling, reference undefined types or functions, or have inconsistent naming across tasks. A self-review for spec coverage, placeholders, and type consistency catches these issues.

Two options exist: subagent-driven execution, which dispatches a fresh subagent per task with review between tasks, or inline execution, which runs tasks in the current session in batches with checkpoints for review.