ghiret/execute

Executing Plans

Overview

Load plan, review critically, execute ALL tasks to completion autonomously.

Core principle: Autonomous execution — run all tasks without stopping for feedback. The user will interrupt if they need to.

Announce at start: "I'm using the execute skill to implement this plan."

Before executing, read the shared conventions:

• _shared/html-conventions.md for HTML vs Markdown plan parsing
• _shared/testing-conventions.md for test speed budgets and slow-test classification

The Process

Step 1: Load and Review Plan

1. Read plan file.
   • For .html plans, parse phases from <section data-phase="N" data-title="...">, tasks from <ul class="tasks"><li>..., and verification from <section class="verification">.
   • For .md plans, parse legacy ## Phase N: Title, - [ ] tasks, and ### Verification sections.
2. Review critically - identify any questions or concerns about the plan
3. If concerns are blocking (cannot proceed without clarification): raise them and stop
4. If concerns are non-blocking (minor, can make a reasonable decision): note them and proceed
5. Create tasks and proceed

Step 2: Execute Batch

Default: First 3 tasks

For each task:

1. Mark as in_progress
2. Follow each step exactly (plan has bite-sized steps)
3. Run verifications as specified
4. Run the tests: unit test, integration tests, frontend tests, hypothesis based tests
5. Fix all failing tests
6. Apply the testing budget: unit tests should be fast, no unit test may exceed 60 seconds, and long integration/e2e tests must be explicitly marked instead of disguised as unit tests.
7. Ask yourself, have I tested this enough? Are the tests that I have written good? Are some of them redundant or too slow? Can I write better, faster tests? And think hard about how to improve the tests, then implement the results of this thinking.
8. Run the tests: unit test, integration tests, frontend tests, hypothesis based tests
9. Fix all failing tests
10. Ask yourself again whether the tests are useful, bounded, and fast. Remove test bloat, sleeps, unbounded generated data, and unnecessary broad snapshots.
11. Run the tests: unit test, integration tests, frontend tests, hypothesis based tests
12. Fix all failing tests
13. Mark as completed

Step 3: Continue to Next Batch

Do NOT stop for feedback. Immediately proceed to the next batch of tasks.

• Log what was implemented for the summary
• Move to the next 3 tasks
• Repeat until ALL tasks are complete

Step 4: Complete Development

After all tasks complete and verified:

1. Final Test Run: Run the full project test suite to ensure no regressions.
2. Handover: Announce: "Implementation complete and verified. Ready for final review."

When to Stop

STOP executing ONLY when:

• Hit an unrecoverable blocker (missing dependency that can't be installed, critical ambiguity)
• Verification fails repeatedly (3+ attempts) on the same issue
• All tasks are complete

Do NOT stop for:

• Minor concerns — make a reasonable decision and continue
• Feedback checkpoints — the user will interrupt if needed
• Asking whether to continue — just continue
• Cleanup questions — keep the plan file, don't ask

Remember

• Run autonomously from start to finish
• Follow plan steps exactly
• Don't skip verifications
• Keep tests bounded and fast; do not create unit tests that can run longer than 60 seconds
• Don't pause for feedback — the user will interrupt if they need to
• Reference skills when plan references them