executing-plans
Structured approach for implementing architect-provided plans through controlled batch execution with review checkpoints - execute in batches (default 3 tasks), verify each step, stop on blockers; don't force through blockers
When & Why to Use This Skill
This Claude skill provides a structured framework for implementing architect-provided plans through controlled batch execution and mandatory review checkpoints. It solves the problem of 'agent drift' and error accumulation in complex tasks by enforcing a disciplined workflow that prioritizes verification and stops immediately at blockers, ensuring that implementation remains strictly aligned with the original design and quality standards.
Use Cases
- Complex Feature Implementation: Breaking down a multi-layered architectural design into manageable 3-task batches to ensure each component is fully functional and verified before proceeding.
- Systematic Code Refactoring: Executing large-scale codebase migrations by following a strict step-by-step plan with built-in safeguards that prevent the propagation of breaking changes.
- Test-Driven Development (TDD) Integration: Managing the development lifecycle by requiring successful test passes for each task batch before moving to the next phase of the implementation plan.
- Risk-Mitigated Deployments: Following multi-stage deployment or configuration playbooks where each checkpoint requires a status report and confirmation before the agent continues execution.
| name | executing-plans |
|---|---|
| description | Structured approach for implementing architect-provided plans through controlled batch execution with review checkpoints - execute in batches (default 3 tasks), verify each step, stop on blockers; don't force through blockers |
| version | 2025-11-04 |
Executing Implementation Plans
Overview
Structured approach for implementing architect-provided plans through controlled batch execution with review checkpoints.
Process Steps
1. Load and Review
- Read the plan completely
- Understand the full scope
- Raise concerns BEFORE starting
- Ask questions about unclear steps
- Verify you have all dependencies
2. Batch Execution
- Execute tasks in batches (default: 3 tasks per batch)
- Complete each task fully before moving to next
- Follow verification steps built into each task
- Commit after each task as specified in plan
3. Checkpoint Reviews
- After each batch completes, report results
- Show what was accomplished
- Report any issues or deviations
- Wait for feedback before proceeding to next batch
4. Completion Protocol
Upon finishing all tasks:
- Verify all tests pass
- Check nothing was skipped
- Transition to finishing-a-development-branch skill to finalize
Critical Safeguards
STOP IMMEDIATELY when encountering:
- Missing dependencies
- Failed tests that can't be resolved
- Unclear instructions
- Blockers of any kind
Don't:
- Force through blockers
- Make assumptions
- Skip verification steps
- Modify plan mid-execution
Do:
- Stop and ask
- Report the blocker clearly
- Wait for guidance
Plan Adherence
- Follow plan steps exactly
- Don't skip verification procedures
- Maintain communication checkpoints
- Treat plan updates as requiring return to review phase
Batch Size
Default: 3 tasks per batch
Adjust based on:
- Task complexity
- Interdependencies
- Your confidence level
- Feedback from reviewer
When to Stop
- Encountering blockers
- Tests failing unexpectedly
- Instructions unclear
- Need architectural decision
- Pattern doesn't match reality
Integration with Other Skills
- writing-plans - Creates the plans you execute
- finishing-a-development-branch - Final verification after all tasks complete
- verification-before-completion - Use before claiming batch is complete