Writing Plans Skill

Writing Plans is a skill from Superpowers (skill source). It runs once you have a spec or clear requirements for a multi-step task, before any code is touched, and produces an implementation plan detailed enough for "an enthusiastic junior engineer with poor taste, no judgement, no project context, and an aversion to testing" (the README's words) to follow.

It usually follows Brainstorming and hands off to an execution skill.

Key Features

• Saved plan: docs/superpowers/plans/YYYY-MM-DD-<feature-name>.md, unless you prefer another location.
• Standard header: goal, architecture, tech stack, the spec it implements, global constraints and a review focus.
• File map first: decide which files are created or changed and what each is responsible for, preferring small focused files.
• Bite-sized steps: each step is one 2 to 5 minute action (write the failing test, run it, implement, run again, commit), grouped into tasks that each end with a testable deliverable.
• No placeholders: "TBD", "add appropriate error handling", "write tests for the above" and "similar to Task N" count as plan failures; code steps need real code blocks.
• Self-review: check spec coverage, scan for placeholders, and confirm names and signatures stay consistent across tasks.
• Handoff: you review the plan and choose subagent-driven or inline execution.

Use Cases

• Any feature too large to hold in one prompt.
• Plans that another agent, or a cheaper model, will execute.
• Work that must survive context compaction or a new session.

Pricing

Free and open source under the MIT license. Prime Radiant, the company behind Superpowers, sells commercial support to enterprises. The practical cost is tokens: process skills add questions, reviews and subagent runs.

Getting Started

Superpowers installs as one plugin, so this skill arrives with the rest of the library (15 skills in v6.4.1). In Claude Code run /plugin install superpowers@claude-plugins-official. Cursor uses /add-plugin superpowers, Gemini CLI uses gemini extensions install https://github.com/obra/superpowers, and the README lists commands for Codex, GitHub Copilot CLI, OpenCode, Pi and other harnesses. Skills trigger on their own; you can also ask for one by name. After a spec is approved, the agent writes the plan and links it for your review.

Limitation: detailed plans take time and tokens to write, and a plan is only as good as the spec behind it. Very small changes are better served by a short design in chat.

FAQ

Why so much detail?

So execution needs no judgment calls the plan could have settled, which lets cheaper models or fresh subagents do the work.

Who executes it?

Subagent-Driven Development or Executing Plans, your choice.

Alternatives

• Planning with Files: Manus-style persistent markdown planning.
• Spec Kit: GitHub's spec-driven toolkit with its own plan and task stages.
• Unlazy: acceptance gates for large tasks.

Conclusion

The bridge between an agreed design and code an agent can build without guessing. More in the skills hub.