README
¶
replay — offline agent loop from a recorded transcript
What this shows
Driving the full agent loop with no LLM credentials by replaying a
recorded JSONL transcript through the mock.NewScripted provider. The
2-turn transcript is inlined as a constant and materialized to a temp
file; in lenient mode (strict=false) the provider plays responses
back in order regardless of the incoming prompt — pass true to
assert each request matches the recording (useful for catching
prompt-construction regressions in tests).
In real usage the transcript comes from a live run captured with the
bundled CLI's --record-to flag or mock.NewRecorder in library code.
Run it
No credentials needed, runs offline:
go run ./examples/replay
Key APIs
mock.NewScripted—github.com/go-steer/core-agent/v2/pkg/models/mockagent.New/agent.Agent.Run—github.com/go-steer/core-agent/v2/pkg/agent
What you should see
user: What is 2+2?
model: Replay turn 1: hello from the transcript.
user: And 3+3?
model: Replay turn 2: this came from the JSONL fixture.
The prompts deliberately don't match the recorded ones — that's lenient mode doing its job.
Next
autonomous— the same scripted provider drivingautonomous.RunAutonomous.with-subagent— two scripted providers, one per agent.- Library guide:
docs/site/src/content/docs/embed/guide.md
Documentation
¶
Overview ¶
Example: drive the agent loop offline by replaying a recorded JSONL transcript through the mock "scripted" provider — no LLM credentials required.
go run ./examples/replay
In real usage the transcript is captured from a live provider with the bundled CLI's --record-to flag (or models/mock.NewRecorder from library code). Here it's inlined as a constant so the example runs standalone.
Source Files