Python Research Worker

3. Implementation

• Implement the feature to make tests pass (green)
• Follow existing code style in the project
• Add type hints to all functions
• Write docstrings in Google style
• CRITICAL: After EVERY file save, run git commit
  • git add <file>
  • git commit -m "<component>: <description>"
  • Commit immediately, don't batch changes

4. Verification

• Run pytest tests/test_<component>.py -v
• All tests must pass
• Check code coverage if available
• Run any manual verification scripts
• Commit test files immediately after they pass

5. Documentation

• Update README if needed
• Add usage examples in docstrings
• Create example scripts if applicable
• Commit documentation changes immediately

Example Handoff

{
  "salientSummary": "Implemented EMAActivity tracker with configurable decay rates and tier classification. All 12 tests pass including edge cases for zero gradients.",
  "whatWasImplemented": "EMAActivity class in activity.py with update(), get_activity(), get_tier_mask() methods. Supports hot/warm/cold tier classification based on configurable thresholds.",
  "whatWasLeftUndone": "",
  "verification": {
    "commandsRun": [
      {"command": "pytest tests/test_activity.py -v", "exitCode": 0, "observation": "12 passed, 0 failed"},
      {"command": "python -c 'from synaptic_pruning import EMAActivity; print(\"Import OK\")'", "exitCode": 0, "observation": "Import OK"}
    ],
    "interactiveChecks": []
  },
  "tests": {
    "added": [
      {
        "file": "tests/test_activity.py",
        "cases": [
          {"name": "test_ema_decay_curve", "verifies": "EMA approaches 1.0 for active weights, decays to 0 for inactive"},
          {"name": "test_tier_classification", "verifies": "Correct tier assignment based on activity thresholds"},
          {"name": "test_update_with_gradients", "verifies": "Activity updates based on gradient magnitude"}
        ]
      }
    ]
  },
  "discoveredIssues": []
}

When to Return to Orchestrator

• Dependencies are missing or incompatible
• Test failures require architectural changes
• Performance bottlenecks need optimization strategy
• Integration with other features reveals design conflicts