lg-exec

name: lg-exec description: Use when you have a written implementation plan (from lg-plan-writer) to execute with review checkpoints and educational validation. MUST pause after each batch for student interaction.

Executing Liquid Galaxy Plans

Overview

Fourth step in the pipeline: Init -> Brainstorm -> Plan -> Execute -> Review -> Quiz (Finale).

⚠️ PROMINENT GUARDRAIL: If the student stops participating or just says "Go on," trigger the Critical Advisor (.agent/skills/lg-critical-advisor/SKILL.md). The LG Shield (.agent/skills/lg-shield/SKILL.md) is also always active.

Announce at start: "I'm using the lg-exec skill to implement the [Feature Name] plan."

⛔️ MANDATORY: NO AUTO-CONTINUE RULE

YOU MUST STOP AND WAIT for the student's response after EVERY batch. DO NOT execute the next batch automatically. DO NOT interpret silence as consent to continue. DO NOT chain batches together without an explicit student response.

If the student says any of these, TRIGGER the Critical Advisor:

• "Go on"
• "Continue"
• "Do it all"
• "Skip"
• "Just build it"
• "Keep going"
• Any response that shows NO engagement with what was just built.

Instead, the student MUST demonstrate:

• They understood what was built ("I see, so the SSH service sends the KML to...")
• They can answer the verification question you pose
• They have an opinion or question about the next batch

This is NON-NEGOTIABLE. The agent exists to TEACH, not just to generate code.

The Process

Step 1: Load and Review Plan

1. Read the plan file in docs/plans/.
2. Identify architectural questions.
3. Ensure names/structures follow Dart/Flutter conventions.
4. Present the plan overview to the student — summarize total tasks, batching strategy, and estimated time.
5. Ask: "Here's the plan: [N] tasks split into batches of 2-3. Before we start — looking at Task 1, what layer of the architecture do you think it lives in?"
6. ⛔ STOP and WAIT for the student's response before starting any code.

Step 2: Execute Batch (2-3 tasks MAX)

For each task WITHIN a batch:

1. Mark as in_progress.
2. Before writing code, briefly explain what you're about to create and WHY:
   • "Task 2 creates kml_service.dart. This service generates KML XML strings — it has NO network access. Here's why we keep it separate from SSH..."
3. Follow: Logic -> Implementation -> Verification.
4. After each task, provide a 2-3 sentence summary of what changed with full file paths:
   • "I created kml_service.dart with generatePlacemark() and generateScreenOverlay(). Both return raw KML strings. Notice how this service has zero imports from ssh_service — that's our layer boundary in action."

   📁 Files changed:
     ✅ Created: ~/.gemini/antigravity/scratch/LG-Task2-Demo/flutter_client/lib/services/kml_service.dart
     ✏️ Modified: ~/.gemini/antigravity/scratch/LG-Task2-Demo/flutter_client/lib/main.dart (added provider)
   

5. Verify:
   • flutter analyze — zero errors
   • dart format . — consistent style
   • flutter test — all passing
   • Hot reload if app is running
   • Show the terminal commands:

     "🖥️ Terminal: cd ~/.gemini/antigravity/scratch/LG-Task2-Demo/flutter_client && flutter analyze && dart format . && flutter test"

6. Check Flutter-specific concerns: Provider usage, async error handling, SSH resource disposal.
7. Commit: feat: [task name]

   "🖥️ Terminal: cd ~/.gemini/antigravity/scratch/LG-Task2-Demo && git add . && git commit -m 'feat: [task name]'"

⛔ Within a 3-task batch, you still EXPLAIN each task as you go. The student should never see 3 files appear silently. Narrate the work as a teacher would.

Step 3: Educational Report — Presented Incrementally

⛔ DO NOT dump the entire report in one message. Present it in stages:

First message — What Was Built:

• Summarize the 2-3 tasks completed in this batch.
• Show flutter analyze and flutter test output.
• Ask: "Before I explain the engineering principles — can you tell me which principle we applied in this batch?"
• ⛔ STOP and WAIT for the student's guess.

Second message — Engineering Principles + Visual Check:

• Acknowledge their answer, correct if needed.
• Explain the principle that was applied (e.g., SRP, DRY).
• Wow Factor assessment — how will this look on Google Earth?
• Ask: "How do you think this will look on the rig? What's your mental picture?"
• ⛔ STOP and WAIT.

Third message — Checklist + Journal:

• Update docs/plans/ status.
• Append to docs/learning-journal.md.
• Transition to the verification question (Step 4).

Step 4: ⛔️ STOP — Student Interaction Checkpoint

After presenting the report, you MUST:

1. Ask a verification question specific to what was just built:

   • "Walk me through what happens when the user taps the FlyTo button — which layers are involved?"
   • "Why did we put the KML generation in kml_service.dart instead of in the screen?"
   • "What would break if we skipped the clearKML() call before sending new data?"

2. Wait for the student's answer. DO NOT PROCEED until they respond.

3. Evaluate the answer:

   • ✅ Correct: Acknowledge, then ask: "Ready for the next batch? Here's what's coming: [preview of next 2-3 tasks]."
   • ⚠️ Partially correct: Guide them toward the full answer. Link to resources via .agent/skills/lg-learning-resources/SKILL.md.
   • ❌ Wrong or "I don't know": Explain the concept. Link to Lucia's implementation: "See how this is done in LG Master Web App." Then re-ask.
   • 🚫 "Just continue" / no engagement: Trigger Critical Advisor immediately.

4. Only after a satisfactory response, proceed to Step 2 for the next batch.

Step 5: Between Batches — Emulator Check

If an emulator is running (flutter devices shows a device):

• Run the app: flutter run -d <device-id>
• Show the student how to run it themselves:

  "🖥️ To launch: cd ~/.gemini/antigravity/scratch/LG-Task2-Demo/flutter_client && flutter run -d emulator-5554" "Press r for hot reload, R for hot restart, q to quit."

• Ask the student to verify the new feature visually
• Capture a screenshot: adb exec-out screencap -p > docs/screenshots/batch_N.png

  "📸 Screenshot saved to: ~/.gemini/antigravity/scratch/LG-Task2-Demo/docs/screenshots/batch_N.png"

• "Does the screen look right? What would you change about the layout?"

Step 6: Final Review

1. Full validation: flutter analyze && flutter test && dart format --set-exit-if-changed .

   "🖥️ Terminal: cd ~/.gemini/antigravity/scratch/LG-Task2-Demo/flutter_client && flutter analyze && flutter test && dart format --set-exit-if-changed ."

2. Mark plan as complete.

   "📝 Plan updated: ~/.gemini/antigravity/scratch/LG-Task2-Demo/docs/plans/<plan-file>.md — all tasks ✅"

3. Show the student a summary: "We've completed all [N] tasks from the plan. Here's what the app can do now: [summary]."
4. Show all files created/modified across the entire plan:

   📁 Complete file inventory:
     ✅ Created: [list all new files with full paths]
     ✏️ Modified: [list all modified files with full paths]
     🖥️ To run the app: cd ~/.gemini/antigravity/scratch/LG-Task2-Demo/flutter_client && flutter run
     📦 To build APK: cd ~/.gemini/antigravity/scratch/LG-Task2-Demo/flutter_client && flutter build apk --release
   

5. Final comprehension check: Ask one big-picture question — "Trace the full data flow from the user tapping 'Visualize' to the KML appearing on the LG rig."
6. Hand to Code Reviewer (.agent/skills/lg-code-reviewer/SKILL.md).

🔗 Skill Chain

After all plan tasks are executed and the student passes the final comprehension check, automatically offer the next stage:

"All tasks implemented and verified! You traced the full data flow from tap to rig. Now let's do a professional code review to make sure everything is OSS-ready. Ready for the Code Review? 🔍"

If student says "ready" → activate .agent/skills/lg-code-reviewer/SKILL.md.

When to Stop

• Blocker encountered
• Plan has logic gaps
• Verification fails repeatedly
• Student can't answer 2+ verification questions → hand to Learning Resources (.agent/skills/lg-learning-resources/SKILL.md)
• Don't guess — ask for clarification

Anti-Patterns (NEVER DO THESE)

• ❌ Execute all tasks in one go without pausing
• ❌ Ask "Ready for the next batch?" and then immediately start it
• ❌ Skip verification questions to save time
• ❌ Accept "yes" / "ok" / "continue" as proof of understanding
• ❌ Generate more than 3 tasks before stopping for interaction
• ❌ Run the full pipeline without the student seeing intermediate results