Executing Plans Skill

Executing Plans is a skill from Superpowers (skill source). It is the inline way to carry out a plan from Writing Plans: the agent implements every task itself in the current session, with no implementer or reviewer subagent per task, and dispatches one fresh-context review of the whole branch at the end.

The skill frames it as the cheaper option: one context plus one final reviewer, instead of a fresh implementer and reviewer for every task.

Key Features

• When to use: you chose inline execution at the plan handoff, or the harness has no subagent tool.
• Continuous execution: no "should I continue?" pauses between tasks.
• Rulings, not stalls: conflicts and plan defects are decided and recorded in a ledger as Ruling: lines that state the decision, the reason, and the cost if it is wrong.
• Four hard stops: irreversible or destructive operations, security-sensitive actions, side effects outside the worktree such as merges or pushes, and a plan too broken to follow.
• Shared workspace: a git-ignored .superpowers/sdd/<plan>/ directory holds the ledger and artifacts, shared with Subagent-Driven Development so executors can be swapped mid-plan.
• Completion contract: a task is done only when its named tests ran and passed in this session, every Expected: line was compared with real output, and every deviation has a ruling. A task-done script records the run.
• Final review: one reviewer, on the most capable model, reads the whole branch.

Use Cases

• Well-specified plans where per-task review would cost more than it saves.
• Harnesses without subagents.
• Running the plan on a mid-tier model, which the skill says works well.

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. At the end of planning, choose inline execution; the agent works through every task and reports at the end.

Limitation: without per-task review, mistakes surface later, and on long plans the last tasks run on the most crowded context. The skill recommends subagent-driven execution for those cases.

FAQ

Does the agent stop to ask me things?

Only for the four hard stops; everything else becomes a recorded ruling.

Can I switch to subagents halfway?

Yes. Both skills share the same ledger format.

Alternatives

• Subagent-Driven Development: per-task subagents and reviews.
• Planning with Files: file-based progress tracking without the review machinery.
• Unlazy: acceptance gates for substantial work.

Conclusion

The budget way to execute a good plan, kept honest by TDD, a ledger and a final review. More in the skills hub.