Gautam-R-Patil/S.O.V.A---OSS---TRUSCOR ? reverse-engineered prompt

Build me the first version of SOVA, a local first Python tool for recording and replaying how an AI agent behaves.

I want a command line app and library that can map what an agent can reach, run controlled tests against it, capture the results as a portable .sova evidence package, and replay that evidence later so I can compare behavior across runs. It should support signed traces, offline verification, deterministic test runs when possible, and safe handling of secrets and redacted data. Add a simple demo workflow that launches a self owned local target, drives it through a real browser, records the trace, and then reproduces it into a report.

Please keep it honest and defensive, if something cannot be verified, mark it as inconclusive instead of guessing. Make it work fully locally, with telemetry off by default, and include enough docs and tests that someone can run the project from a fresh clone. If you need current docs for any browser or tooling pieces, look them up online.