Evals

Traces, artifacts, and files

Inspect raw evidence, captured outputs, file paths, traces, and artifacts.

Trace records, captured files, stdout, stderr, generated outputs, and file evidence show the raw evidence behind a result.

Use Results to identify the row that matters. Use Runs and jobs to find run and job ids.

Show evidence

Use workbench eval show to inspect Eval, run, job, trace, artifact, and file evidence:

bash
workbench eval show RUN_IDworkbench eval show JOB_IDworkbench eval show eval-v2workbench eval show eval-v2:cases/investor-focus/case.yamlworkbench eval show RUN_ID:cases/investor-focus/output/result.json

If a suffix is ambiguous, Workbench prints exact workbench eval show REF:PATH commands. Use eval-vN refs to inspect the exact historical eval definition that produced a result.

Source evidence is separate

Sources hold evidence ingested from local agent sessions or other adapters. They are deployment-owned corpora, not project execution traces, and do not enter a Skill object pack automatically. The Sources workspace renders exact generic citation blocks from immutable snapshots; Eval execution evidence on this page continues to describe only runs and jobs.

Inspection paths

Workbench shows stable inspection paths for evidence files. Runtime paths under .workbench/** and raw trace metadata files are hidden.

For generated files, put outputs and diagnostics under /workspace/output. Keep file-format checks in cases, tests, grading criteria, scripts, or skill references. See File formats for Office, PDF, and similar artifacts.

Browser evidence

The browser UI reads the same inspection data as CLI commands. It can show source files, run details, job timelines, and output files without changing project state.