| name | executing-plans |
| description | Use when you have an approved implementation plan and want to execute it directly in the current session without the full subagent-driven workflow. |
Executing Plans
Use this skill to execute a written implementation plan directly in the current session.
Relationship To Context Skills
When work is tracked, use context-task as the durable execution-record policy and pi-task tools for task-record operations:
- use
task_read for the brief, plan, and prior notes
- use
task_append_note for task-specific discoveries and progress
- use
task_write (document: "plan") when a material course correction changes the coherent plan
- use
task_write (document: "task") for stable task metadata or relevant PR references
Use context-memory only for durable cross-task facts or decisions.
Why This Skill Exists
This is the lighter-weight alternative to subagent-driven-development: execute a plan in one session without per-task builder and reviewer dispatches. Use the subagent workflow when stronger context isolation and review gates are needed.
When to Use This Skill
Use it when the plan is approved, clear enough to execute directly, and moderate in size or tightly coupled. Do not use it when the work is highly parallelizable, benefits from strict per-task isolation, or needs repeated review gates.
Workflow
1. Load and review the plan
Before editing, read the plan carefully and check for contradictions, missing prerequisites, or unclear steps. For tracked work, read the task brief, plan, and relevant notes with task_read. Do not treat a plan as automatically correct just because it exists.
2. Track execution
Identify each plan task, keep one task in progress at a time, and regard a plan task as complete only after its verification passes. For tracked work, record meaningful progress and discoveries with task_append_note rather than relying on transient chat history.
3. Execute one task at a time
For each task:
- understand the scope
- perform the listed implementation steps
- run its verification
- confirm the result before moving on
Record material discoveries and revise the full plan with task_write (document: "plan") when needed. Do not skip ahead merely because later work seems related.
4. Handle problems explicitly
Stop and reassess if the plan is missing detail, verification repeatedly fails, the codebase conflicts with the plan, or a task reveals that the plan structure is wrong. Ask for clarification or revise the plan instead of forcing through blockers.
5. Final validation
After implementation tasks are verified, run broader relevant checks, confirm the implementation still meets the plan goal, and request review when substantial or risky.
Do not call task_finish merely because all implementation tasks are complete. Finish is a separate explicit-user-directed workflow through finishing-task.
Red Flags
Do not:
- start implementation before reviewing the plan
- skip task-level verification
- continue through repeated failures without reassessing
- silently change the plan's intent
- manipulate task files or state directly
- finish a tracked task without explicit user direction
Bottom Line
Review first, execute task by task, verify continuously, and retain durable task-specific progress without automatically changing task lifecycle state.