migrate-gepa_skill

Overview

This skill helps migrate DSPy GEPA usage to gepa-observable to add integrated observability, a real-time dashboard, and LM call logging. It preserves existing GEPA API and workflow while enabling server reporting, console logging, and custom observer callbacks for full lifecycle visibility. Use it to track prompt evolution, review iterations, and capture lineage and LM calls during optimization.

How this skill works

Replace the original GEPA import with gepa_observable.GEPA and pass observability parameters to the constructor. The package wires in multiple observers (server, logging, LM call logger, plus your custom observers) that receive the same eight lifecycle events. No other DSPy GEPA APIs change: metrics, program types, data formats, budget parameters, and compile() remain compatible.

When to use it

• You want a web dashboard to monitor GEPA optimization in real time.
• Teams need to review each iteration and the prompt lineage for auditability.
• You must capture every LM invocation for debugging or cost analysis.
• You want to add custom callbacks without modifying the optimization loop.
• Migrating notebooks or scripts that already use DSPy GEPA with minimal code changes.

Best practices

• Install the package in the same environment as your DSPy runs: pip install dspy-gepa-logger.
• Set server_url for dashboard integration and capture_lm_calls for full LM logging when debugging or auditing.
• Keep verbose=True during development for LoggingObserver summaries, then reduce verbosity in production.
• Implement lightweight custom observers to avoid adding latency to optimization iterations.
• Run notebook cells in order after changing the import; API is compatible so no other code edits are required.

Example use cases

• Add a ServerObserver to display optimization progress and prompt proposals on a shared dashboard.
• Capture and store LM calls during prompt optimization for cost tracking and prompt-level performance analysis.
• Attach a custom observer that saves proposed prompt changes and acceptance decisions to your issue tracker.
• Use LoggingObserver in CI to produce concise iteration summaries for automated regression checks.
• Migrate team notebooks to the observable GEPA to enable collaborative reviews of prompt lineage.