Executing Plans
Turn a written implementation plan into verified, checkpointed development work.
The skill requires critical plan review, human confirmation on concerns, stopping at blockers, isolated worktrees, and explicit consent before working on main/master. It does not specify sensitive-data handling, external side effects, rollback, or transparent data-flow controls; recovery is delegated to unspecified sub-skills, so 8 points are deducted.
The sequence, blocker conditions, and verification expectations are clear, and failures should be explained rather than guessed through. However, execution depends on a plan file, task tracking, and several sub-skills; dependency availability, diagnostic formats, and key-path tests are not supplied. The static cap supports 9 points.
The target scenario is clear: executing an existing written implementation plan with review checkpoints in coding-agent environments. Input-plan structure, non-fit cases, varying agent capabilities, and Chinese-language use are not well bounded; some functionality also depends on platform or sub-agent support, so 5 points are deducted.
The document is concise and layered, with activation context, process steps, stop rules, and required integrations. Repository context supplies an MIT license, version, and general maintenance/update information. The selected file lacks parameters, examples, FAQs, known limitations, changelog details, and explicit maintenance ownership, so 7 points are deducted.
Following bite-sized plan tasks, verification steps, and completion handoff can address the core execution goal. Output format, plan-quality gates, and concrete deliverables are unspecified; the result depends on other skills and there is no representative static output or execution evidence. The static cap supports 6 points.
The process requires plan-specified verification and emphasizes evidence before completion. Repository context mentions tests and an evaluation harness, but no dedicated tests for this skill's key path or independently reproducible results are provided here. Evidence is therefore limited, supporting 4 points under the static cap.
- Before execution, confirm that the plan file, task tracking, and all required sub-skills are available; otherwise the workflow may block mid-run.
- The skill permits code changes but provides no least-privilege or confirmation checklist for sensitive files, network calls, credentials, or external side effects.
- Completion verification and branch merge, retention, or discard recovery paths depend on finishing-a-development-branch and should be reviewed separately.
- No Chinese-language guidance is provided, and reachability of platform-dependent plugins, sub-agents, or external services from mainland-China networks is unverified.
What does this skill do, and when should you use it?
Executing Plans is for running an existing written implementation plan in a separate session. It first reads and critically reviews the plan, stopping to raise concerns before implementation begins. Once approved, it executes tasks one by one and runs the specified verification steps. After completion, it requires the finishing-a-development-branch workflow; blockers, repeated verification failures, and unclear instructions require a pause for help.
Reads and critically reviews an implementation-plan file; creates todos for plan items; marks each task in progress or completed; follows the plan's steps; runs the verifications specified by the plan; stops and asks for clarification when dependencies are missing, instructions are unclear, or verification fails repeatedly; invokes finishing-a-development-branch after all tasks are complete.
- A software engineer has a detailed implementation plan and wants an agent to execute it in a separate session with human checkpoints.
- A team wants plan concerns surfaced before coding begins rather than having the agent guess through gaps.
- A project has bite-sized tasks with explicit verification steps that should be tracked through completion.
- A user is working on a platform with subagent support and wants to evaluate whether the recommended subagent-driven workflow should replace plan execution.
What are this skill's strengths and limitations?
- Reviews the plan before implementation and surfaces critical concerns early.
- Tracks tasks individually and requires the plan's verification steps.
- Defines explicit stop conditions for blockers, unclear instructions, and repeated verification failures.
- Fits separate-session execution with human review checkpoints.
- Requires a written implementation plan and does not replace plan creation.
- Requires the `finishing-a-development-branch` workflow after implementation and relies on related Superpowers skills.
- The source recommends `subagent-driven-development` when subagents are available, so this may not be the preferred workflow on those platforms.
- The skill source does not provide a standalone install command or a dedicated test suite.
How do you install this skill?
Install the Superpowers repository as a collection; this skill is located at skills/executing-plans/SKILL.md. Follow the README instructions for your harness. For Codex CLI, open /plugins, search for superpowers, and select Install Plugin; in Codex App, install Superpowers from the Plugins sidebar under Coding. The source does not document a standalone installation command for this skill.
How do you use this skill?
Provide a written implementation plan and ask the agent to execute it in a separate session. Example: "Read this implementation plan, review it for concerns first, then execute each task with the specified verification and pause at review checkpoints." When subagents are available, the source recommends using subagent-driven-development instead.
How does this skill compare with similar options?
Compared with subagent-driven-development, this skill executes a written plan with checkpoints, while the README describes the alternative as dispatching a fresh subagent per task with two-stage review. The source recommends the subagent-driven workflow when subagents are available.