| name | execute-plan |
| description | Execute a written implementation plan in this session with checkpoints. |
| disable-model-invocation | true |
You are running the execute-plan skill. This is manual inline execution of a written plan.
If the skill was invoked with arguments, treat them as the plan path. Otherwise, identify the correct plan file before doing anything else.
Goal
Load a plan, review it critically, execute it in this session, and stop when blocked instead of guessing.
Before You Start
- If work should be isolated and you are not already in an appropriate branch or worktree, suggest or run the setup-worktree skill first.
- Never begin implementation on
main or master without explicit user consent.
- Read the entire plan before touching code.
Phase 1: Review the Plan
Before starting execution:
- read the plan end to end
- identify missing information, contradictions, or risky assumptions
- raise concerns before starting any task
If the plan has critical gaps, stop and ask the user to update the plan instead of improvising around it.
Phase 2: Create Execution Tracking
Maintain an explicit checklist in the conversation with one entry per plan task.
Rules:
- only one task should be
in_progress at a time
- mark tasks complete immediately after verification succeeds
- add follow-up items only when they are genuinely required by the plan or by issues discovered during execution
Phase 3: Execute Tasks Exactly
For each task:
- Mark it
in_progress.
- Follow each step in order.
- Run the verification commands the plan specifies.
- If a step fails, stop and resolve the failure before moving on.
- Mark the task complete only after verification succeeds.
Do not silently skip steps because they feel repetitive.
TDD Rules
Implementation work in the plan must follow TDD:
- write the failing test first
- run it and confirm it fails for the expected reason
- write the minimal implementation
- rerun the test and confirm it passes
- refactor only while keeping tests green
If code was written before the failing test, acknowledge that and restart the step properly instead of pretending it was test-driven.
Verification Rules
Do not claim a task is complete without fresh evidence.
Before saying something passes, is fixed, or is complete:
- identify the command that proves the claim
- run that command now
- read the output and exit status
- report the actual result, not the hoped-for result
Partial checks are not enough when the plan calls for broader verification.
When to Stop and Ask
Stop immediately when:
- the plan is unclear
- a verification step fails repeatedly
- a dependency or environment issue blocks progress
- the required behavior conflicts with reality in the codebase
- you no longer understand why you are making a change
Do not push through blockers by guessing.
Handoff
After all tasks are complete and verified, run the finish-development-branch skill.
If the user would benefit from higher-quality task isolation and review gates, recommend the execute-plan-with-subagents skill instead of continuing inline.