atomic_execution

Atomic Execution Workflow

Use this workflow for every coding task to ensure quality and prevent "Agent Hallucinations".

The Loop

1. Discuss (Review)
2. Plan
3. Execute
4. Verify

---

1. Discuss (Review)

Before writing code, understand the context.

• Read .agent/STATE.md.
• Read relevant source files.
• Ask clarifying questions if requirements are vague.

2. Plan (The Blueprints)

Create a specific plan for this specific task. Do not generic plan.

• For complex tasks: Create/Update .agent/plans/current_task.md.
• For simple tasks: Output a markdown checklist in the chat.

Checklist Format:

- [ ] Create `file_a.py` with function `x`.
- [ ] Update `file_b.py` to import `x`.
- [ ] Run test `test_x.py` to verify.

3. Execute (Atomic Steps)

Write code in small, verifiable chunks.

• Do not write 5 files at once.
• Do write one module, then the next.
• Do use edit_file or write_file tools precisely.
• Rule: If you edit a file, you must know its current content (read it first).

4. Verify (Trust, but Verify)

Never claim a task is done without proof.

• Compile/Lint: Run the build command.
• Test: Run specific unit tests (pytest tests/test_feature.py).
• Manual Check: If UI, verify the component renders (conceptually or via screenshots if available).

Completion

Once verified:

1. Update .agent/STATE.md.
2. Mark the item as done in .agent/ROADMAP.md (if applicable).
3. Notify the user.