Executing Plans
Overview
Load plan, review critically, execute tasks in batches, report for review between batches.
Core principle: Batch execution with checkpoints for architect review.
Announce at start: "I'm using the executing-plans skill to implement this plan. REQUIRED: Ensure Antigravity is set to Fast Mode for efficient execution."
Step 0: Prerequisites
STOP if:
- (a) You are on main or master branch without explicit user instruction
- (b) There are uncommitted changes not related to this plan
Resolve both conditions before proceeding. Never start implementation on main/master branch without explicit user instruction.
The Process
Step 1: Load and Review Plan
- Read plan file
- Review critically - identify any questions or concerns about the plan
- If concerns: Raise them with your human partner before starting
- If no concerns: Create TodoWrite and proceed
Step 2: Execute Batch
Batch size criteria:
- Default batch: 3 tasks
- Shrink to 1 task if: first task in a new phase, or previous batch had failures
- Expand to 5 tasks if: all tasks are mechanical (file copies, config changes) with no logic
For each task:
- Mark as in_progress
- Follow each step exactly (plan has bite-sized steps)
- Run verifications as specified
- Mark as completed
Step 3: Report
When batch complete, use this template:
## Batch Report
Tasks completed: [N-M]
Files changed: [list]
Tests: [passed X / failed Y / skipped Z]
Verification output:
[paste exact command output]
Ready for feedback.
If Tests: failed > 0: Do NOT mark tasks completed. Report the failure with exact test output. Load systematic-debugging skill. Do not proceed to next batch until failures are resolved.
Step 4: Continue
Based on feedback:
- Apply changes if needed
- Execute next batch
- Repeat until complete
Step 5: Complete Development
After all tasks complete and verified:
- Announce: "I'm using the finishing-a-development-branch skill to complete this work."
- Before loading: verify (a) all tasks are marked completed, (b) all tests pass, (c) no BLOCKERS.md is active.
- REQUIRED SUB-SKILL: Load finishing-a-development-branch
- Follow that skill to verify tests, present options, execute choice
When to Stop and Ask for Help
STOP executing immediately when:
- Hit a blocker mid-batch (missing dependency, test fails, instruction unclear)
- Plan has critical gaps preventing starting
- You don't understand an instruction
- Verification fails repeatedly
Ask for clarification rather than guessing.
If the session ends without human response, or if 2 or more messages have passed without acknowledgment of the STOP:
- Document the blocking question in a
BLOCKERS.mdfile - Mark the plan task as 'blocked'
- Skip to the next non-dependent task if one exists
- If no tasks are unblocked, leave a clear status note and stop
When to Revisit Earlier Steps
Return to Review (Step 1) when:
- Partner updates the plan based on your feedback
- Fundamental approach needs rethinking
Don't force through blockers - stop and ask.
Remember
- Complete Step 0 prerequisites before anything else
- Review plan critically first
- Follow plan steps exactly
- Don't skip verifications
- Reference skills when plan says to
- Between batches: just report and wait
- Stop when blocked, don't guess
Integration
Required workflow skills:
- writing-plans - Creates the plan this skill executes
- finishing-a-development-branch - Complete development after all tasks
Completion State
The session is complete when:
- All plan tasks are marked completed
- Final batch report shows zero test failures
- finishing-a-development-branch skill has been loaded and executed
Announce: "Plan execution complete. All [N] tasks finished. Branch ready for review."