feat(api): add avg tool calls and avg model calls

[Peanut-Puff]

Summary by CodeRabbit

• New Features
  • Added trajectory statistics (average tool calls and model calls) to job summary display
  • Enabled visualization of nested subagent trajectories within observation results in the trajectory viewer

[coderabbitai]

Caution

Review failed

The pull request is closed.

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: f62cb26f-ea87-4bea-9985-6a192184ebc7

📥 Commits

Reviewing files that changed from the base of the PR and between d43f13f and 09efe55.

📒 Files selected for processing (7)

• apps/viewer/app/lib/api.ts
• apps/viewer/app/lib/types.ts
• apps/viewer/app/routes/job.tsx
• apps/viewer/app/routes/trial.tsx
• src/harbor/agents/installed/bitfun_cli.py
• src/harbor/viewer/server.py
• tests/unit/agents/installed/test_bitfun_cli.py

---

📝 Walkthrough

Walkthrough

The PR introduces support for tracking and displaying subagent trajectory relationships. It extends the data model to include subagent references and metadata provenance, backfills those references from child session metadata on the backend, aggregates trajectory statistics, and displays both metrics and embedded subagent traces throughout the frontend.

Changes

Subagent Trajectory References and Metrics

Layer / File(s)	Summary
Data Model Extensions for Subagent References apps/viewer/app/lib/types.ts	SubagentTrajectoryRef interface is introduced with optional trajectory, session, and path identifiers plus metadata. ToolCall, Step, TrajectoryAgent, and ObservationResult each gain optional extra metadata fields. ObservationResult adds optional subagent_trajectory_ref[] array. Trajectory adds optional trajectory_id and recursive subagent_trajectories[] fields.
Backend Metadata Backfill for Subagent References src/harbor/agents/installed/bitfun_cli.py, tests/unit/agents/installed/test_bitfun_cli.py	_attach_subagent_refs_from_metadata() scans child subagent session metadata files, extracts parent tool-call identifiers from relationship fields, matches those identifiers to parent session steps, and populates SubagentTrajectoryRef entries in observation results. Deduplicates against already-referenced subagent ids and annotates both parent Step.extra and inserted refs with provenance flags. Unit test validates backfill behavior.
Backend Trajectory Statistics Aggregation src/harbor/viewer/server.py	New GET endpoint /api/jobs/{job_name}/trajectory-stats scans trial directories for agent/trajectory.json files, aggregates tool-call and model-call counts across trajectory steps, and returns per-job trajectory count and average metrics (rounded to 1 decimal place) or None when no valid trajectories exist.
Frontend Job Page Statistics Display apps/viewer/app/lib/api.ts, apps/viewer/app/routes/job.tsx	TrajectoryStats client-side type and fetchTrajectoryStats() function call the backend endpoint. Job page fetches trajectory stats via React Query and conditionally renders avg_tool_calls and avg_model_calls metrics in the job header row.
Frontend Subagent Trace Rendering in Trials apps/viewer/app/routes/trial.tsx	Adds findSubagentTrajectory() helper, SubagentTraceList accordion component, and SubagentTrace component to resolve and render embedded subagent traces within trial steps. Updates StepContent to accept optional subagentTrajectories prop and rewires observation result rendering to include subagent trace lists alongside content renderers.

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~25 minutes

Possibly related PRs

• JinnanDuan/bitfun-harbor#4: Includes the same subagent metadata backfill logic and corresponding type/rendering/test updates for subagent trajectory references.

Poem

🐰 A rabbit hops through trajectories deep,
Finding lost trails where subagents creep.
With metadata whispers and stats to compile,
The viewer now shows every nested beguile! 🌿

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}