SLM_Agent

Evaluate the current codebase for ScaleDown SLM integration opportunities. Scans all Python and TypeScript/JavaScript files, traces the full purpose of each AI API call (following imports and helper functions across files), classifies each call using your own judgment, scores complexity and suggests decomposition for multi-task calls, generates a structured migration plan, and saves it as scaledown-report.md.

Optimize the prompt you pass to a ScaleDown SLM (extract, classify, summarize, or compress) and benchmark it against a baseline — the frontier-model output it replaced, or your ground truth — on the user's own data. Collects sample data and a ScaleDown API key; if the user has no samples, walks them through creating and validating a small labeled set first. Then runs repeatable evals with a metric appropriate to the task type, quantifies run-to-run noise, and iterates the prompt until it beats — or matches within noise — the baseline. General by design: adapts to any use case, with task-specific playbooks.