Skip to content

Add session diff mode #11

Description

@lewing

Feature: Diff Mode

replay diff session-a.json session-b.json

Compare two eval runs or transcripts side-by-side. Show where tool call sequences diverge, which turns differ, and highlight the critical decision point where one model succeeded and another failed.

Motivation

When analyzing Arena progression results across models (e.g., GPT-5.1-Codex vs Sonnet 4 on the same eval prompt), the #1 manual task is figuring out what the weaker model did differently. Currently this means opening two JSON files and reading them in parallel.

Suggested Output

  • Turn-by-turn alignment showing matching vs divergent turns
  • Tool call diff (which tools were called, in what order)
  • Summary: "Diverged at turn 5: Model A called maestro_subscriptions, Model B tried gh CLI"

Use Cases

  • Cross-model eval comparison in Arena progression pipeline
  • Before/after comparison when a skill prompt changes
  • Regression hunting: "what changed between v1 and v2 of this skill?"

Related

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    enhancementNew feature or request

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions