Pre-Flight Run Forecast

“Before you run AgentSmack — know the ~probes, ~tokens, ~cost, and ~latency.”

The Test-Plan Synthesizer plans which deep-test surfaces to run and why; this is its companion — it forecasts the run’s resource envelope before any probe fires. Paste an agent definition — system prompt + declared tools + framework + model — and it projects, deterministically: how many probes will land (per category), across how many recommended deep surfaces, the token budget, an estimated USD cost band, and a latency band (a static per-probe / per-turn projection, never a measured time). It reuses the very same canonical probe selector and test-plan synthesizer the live run uses, so the estimate is honest, not invented. Honest-empty: a bare prompt with no tools yields a minimal forecast — zero deep surfaces, never a fabricated estimate. Raw prompt bytes never enter the forecast. Load a sample to watch the envelope swing with no live infra.