Smart Planning

Phase 2: Technical Planning (The Blueprint)

Create a strict, bite-sized implementation plan at docs/plans/YYYY-MM-DD-<feature-name>.md. Assume the executing session has zero context. Write exact file paths, exact code blocks to change, and exact commands to run.

Document Header Format:

# [Feature Name] Implementation Plan

**Goal:** [One sentence what this builds]
**Architecture:** [2-3 sentences]

Task Structure (Bite-Sized):

### Task 1: [Component Name]
**Files:**
- Create: `exact/path/to/file.py`
- Modify: `exact/path/to/existing.py:123-145`

**Step 1: Write/Update Code** (Provide exact diff or function replacement)
**Step 2: Verification** (Provide exact command: `npm run build` or `pytest ...`)
**Step 3: Commit** (Command: `git commit -am "feat: added X"`)

Note: DRY, YAGNI, TDD principles apply.

Phase 3: Batch Execution (Action)

Once the plan is written (and approved by the user if it's a major change):

1. Auto-Checkpoint: RUN git add . && git commit -m "chore: auto-checkpoint before <feature>" IF the working tree is dirty. (Safety first!)
2. Execute in Batches: Do not execute 10 tasks at once. Process 2-4 tasks strictly, verifying each step as written in the plan.
3. Report: Stop execution, summarize what was achieved (e.g. "Baigta 1 Partija (UI struktūra)"), present the result to the user, and ask for permission to proceed to the next batch.

When to Stop and Escalate

The Agent MUST stop and notify the user immediately if:

1. The 3-Strike Error Protocol fails: You tried to fix an error 3 different ways during execution and it still fails. Do not guess. Do not loop. Escalate.
2. Missing Dependencies: The plan relies on a tool/library that doesn't exist and you don't have permission to install it.
3. Ambiguous Instructions: The user's prompt contradicts existing rules in gemini.md.

Summary of Deprecated Skills

This skill completely replaces and obsoletes:

• planning-with-files
• writing-plans
• executing-plans Do not use them. Use smart-planning.