pytracer-replay

Layer 4 of pyTracer: reconstructs a past agent run deterministically, injects recorded tool/model responses instead of live calls, runs assertions, diffs two runs, and classifies a cross-branch reproduction as fixed / still failing / diverged.

from pytracer_replay import MockSet, ReplayEngine, ReplayMode, classify
import pytracer_sdk as pytracer

# 1. Build a mock set from a captured trace (its recorded tool/model outputs).
mocks = MockSet.from_trace(recorded_spans)

# 2. Replay the agent deterministically — tools and model are mocked.
engine = ReplayEngine(mode=ReplayMode.FULL)
result = engine.replay(lambda: agent_entrypoint(task), mocks)

# 3. Assert, diff, fingerprint, or classify against another branch's run.

Modes: FULL (recorded model + tools), PARTIAL (recorded tools, live model), LIVE (both live, for drift detection).