feat(usage): report application and judge costs separately

[sentry-junior]

Closes #102.

Preserve every judge-model run and report its usage separately from the application under test.

What changed:

• UsageSummary.costUsd is a typed, optional, non-negative USD value; omit it when unknown and use 0 only for known-free runs.
• Judge calls run through judgeHarness + ctx.runJudge(...); each normalized run is retained under its score's judgeRuns.
• GitHub, terminal, and local report UI show application, judge, and combined tokens and cost.
• Partial costs are never presented as complete totals, including across cases and shards.
• AI SDK, OpenAI Agents, and Pi judge adapters preserve normalized usage; Pi maps its native total USD cost to costUsd.
• Persisted report workspaces use schema v2. Raw Vitest task.meta readers and producers must upgrade together.
• Public docs, llms.txt, repo instructions, examples, and the bundled agent skill use session.events and the judge-harness pattern.

Verification:

• typecheck, lint, docs check, docs build
• 99 reporting/UI tests plus 34 judge/reporter tests
• full suite passes except the known root-permission test (chmod cannot make the fixture unreadable as root)
• browser QA at 1440×900 confirmed workspace and case usage breakdowns

via David Cramer.

--

View Junior Session [Sentry]

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
vitest-evals-docs	Ready	Preview	Sep 10, 2026 1:49pm UTC

[cursor]

Stale Bugbot comment from a previous run.

[cursor]

Stale Bugbot comment from a previous run.

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 496a008. Configure here.}