feat: Demo enhancement, Git sync command, ML/Keras agents

[RonRichman]

Overview

This issue addresses four improvements to the compound-eng workflow system:

1. Demo Command Enhancement - Integrate multi-llm-brainstorm and add notebook API validation
2. Notebook API Validation Agent - Create specialized agent to ensure notebook API correctness
3. Git Sync Command - New slash command to reset local to master and merge with remote
4. ML/Keras Agents - Specialized agents for machine learning code review

---

Issue 1: Demo Command Enhancement

Problem

The /demo command creates Jupyter notebooks but doesn't leverage the multi-llm-brainstorm agent for diverse perspectives. Additionally, notebooks sometimes contain outdated API calls that fail on execution.

Solution

• Always invoke multi-llm-brainstorm for demo strategy planning
• Add mandatory notebook API validation step before finalization
• Invoke new notebook-api-validator agent

Acceptance Criteria

• /demo always invokes multi-llm-brainstorm for strategy consultation
• API validation runs before notebook finalization
• Notebooks verified to execute without API errors

---

Issue 2: Notebook API Validator Agent

Problem

Jupyter notebooks often contain outdated API calls due to rapid library updates, deprecated methods, and parameter changes.

Solution

Create notebook-api-validator agent that:

• Parses notebook code cells for external library usage
• Fetches current API documentation via Context7 MCP
• Validates all API calls against current signatures
• Reports issues with specific fixes

Files to Create

• agents/domain/notebook-api-validator.md

---

Issue 3: Git Sync Command

Problem

Users need a simple way to reset their local branch to match remote master.

Solution

Create /git-sync-master command that:

• Stashes any local changes (with confirmation)
• Fetches latest from remote
• Resets local to remote master
• Optionally restores stashed changes

Files to Create

• commands/git-sync-master.md

---

Issue 4: ML/Keras Agents

Problem

The workflow system lacks specialized agents for ML code, particularly Keras 3 which supports TensorFlow, PyTorch, and JAX backends.

Solution

Create:

• keras3-code-reviewer - Reviews Keras 3 code for backend-agnostic ops, custom layers, serialization
• ml-research-reviewer - Reviews ML implementations, architectures, training procedures

Files to Create

• agents/review/keras3-code-reviewer.md
• agents/review/ml-research-reviewer.md

---

Implementation Order

1. Notebook API Validator Agent (dependency for Issue 1)
2. Demo Command Enhancement
3. Git Sync Command
4. ML/Keras Agents

Files Summary

Create:

• agents/domain/notebook-api-validator.md
• commands/git-sync-master.md
• agents/review/keras3-code-reviewer.md
• agents/review/ml-research-reviewer.md

Modify:

• commands/workflows/demo.md
• local_manifest.yml

---

Full detailed plan: plans/multi-feature-plan-demo-git-ml.md

🤖 Generated with Claude Code