orca-replay - Inspect, replay, and compare agent runs
Answers questions about past agent runs from recordings, replays or forks them, and compares model outcomes.
Tags
Updated: 2026-09-30Capabilities
Typical Inputs
Typical Outputs
What this skill does
- Inspect recorded agent runs
- Trace causal event chains
- Separate recorded and inferred evidence
- Replay recorded runs
- Fork runs across models
- Compare model outcomes
Inputs
- User question
- Agent run recordings
- MCP server
- Model identifiers
- Verification command
Outputs
- Trace-based explanations
- Replay divergence reports
- Fork comparison results
- Run status updates
Requirements
- Node.js 20 or newer
- orcareplay npm package
- Registered orca MCP server
- At least one .orca/runs recording
- Access to supported models