tool-call-replay 0.1.0

Deterministic replay harness for agent tool-call traces.

tool-call-replay

tool-call-replay turns agent tool-call traces into deterministic local replay artifacts and pytest regression tests. It helps agent teams convert production failures into reviewable test cases without live model calls or live tool execution.

Quickstart

python -m venv .venv
. .venv/bin/activate
pip install tool-call-replay
tool-call-replay ingest examples/failed_refund_agent_trace.jsonl --out replay.json
tool-call-replay run replay.json --assert examples/refund_assertions.yaml
tool-call-replay pytest replay.json --assert examples/refund_assertions.yaml --out test_refund_agent_replay.py

Supported Inputs

• simple JSONL traces;
• OpenAI-style event traces;
• OTLP/GenAI span exports;
• Phoenix-like JSON exports.

The v1 importers normalize all inputs into a canonical ToolCallReplay JSON object.

What This Is Not

This is not an agent framework and does not call models, tools, or provider APIs. It replays recorded tool-call structure and mocked outputs so failures can become deterministic tests.