[ndycode]

Summary

• Request Claude stream-json output in the live WorkerSpawner path, route it through the shared parser, and retain bounded human-readable assistant/diagnostic tails while only detecting PR URLs in parsed assistant text.
• Record one classified live-worker cost entry per completed call. Claude's latest total_cost_usd is treated as a client-side estimate and used once when available; Relay's token-price table is only the fallback, so the two sources are never added together.
• Preserve failure, cancellation, partial-output, stderr, and worktree-cleanup behavior; reject unsafe numeric/output sizes; document the observational (not billing-authoritative) cost semantics.

Closes #243

Test plan

• pnpm test (1,248 passed, 29 skipped)
• pnpm typecheck
• pnpm build
• pnpm format:check
• cargo check --workspace --locked
• cargo test -p harness-data (91 passed)
• cd gui && pnpm build
• HARNESS_LIVE=1 ./node_modules/.bin/vitest run test/integration/live-worker-cost.test.ts (real autonomous handler -> Claude -> cost ledger -> rly cost)
• Focused fake-worker and regression coverage for parsed assistant/result output, PR URL detection, malformed and non-JSON diagnostics, partial/oversized output, missing/unsafe usage, provider-only cost, failures, stop behavior, task classification, and cost-source precedence

Size check

Diff size: 1004 lines added / 95 deleted.

This exceeds 800 lines because 724 additions are focused regression and opt-in integration tests. The production/documentation change is 280 additions and 66 deletions across the existing worker, parser, cost-ledger, ticket-classification, and mirrored type paths; it does not introduce a parallel parsing or orchestration framework.

Breaking changes?

No. The new ticket taskType field is optional and the Rust mirror uses a default for existing serialized tickets. Existing non-JSON diagnostics and failure/cleanup behavior remain supported.

AI-assisted?

Yes. OpenAI Codex provided substantial assistance with investigation, implementation, tests, and review. Every commit includes the required Co-Authored-By: OpenAI Codex <noreply@openai.com> footer. The final diff was reviewed and validated at commit 666f4826b31dd1ea675475ff0ac8032609c4c7fd.

[qodo-code-review]

Qodo reviews are paused for this user.

Troubleshooting steps vary by plan Learn more →

On a Teams plan?
Reviews resume once this user has a paid seat and their Git account is linked in Qodo.
Link Git account →

Using GitHub Enterprise Server, GitLab Self-Managed, or Bitbucket Data Center?
These require an Enterprise plan - Contact us
Contact us →