[ramantehlan]

Summary

• Report a server or core log to Sentry only when the catch site marks it with reportable_error.
• Mark infrastructure and unexpected internal failures, including network and timeout errors on mixed authorization and LLM logs. Provider inference errors, bad credentials, rate limits, and client disconnects stay in the logs.
• Stop reporting missing sandbox versions, rejected sandbox credentials, and the duplicate lower-level build failure.

Test plan

• pnpm --filter @truefoundry/trueforge exec jest --config jest.unit.config.cjs tests/unit/sentry/reportMarkedLog.test.ts tests/unit/apis/auth.test.ts tests/unit/sandbox/progressSandboxEnvironmentVersion.test.ts tests/unit/controller/sandboxEnvBuild.test.ts
• pnpm --filter @truefoundry/trueforge-core exec jest --config jest.config.cjs tests/core/llm/VercelAILLM.stream.test.ts
• Confirm a marked log reaches Sentry and an unmarked provider error does not
• Confirm the printed log does not include reportable_error

---

Note

Medium Risk
Broad changes to error observability and many failure paths; mis-tagged reportable_error or isReportableError edge cases could over- or under-alert Sentry without affecting request handling.

Overview
Introduces opt-in Sentry reporting via Winston: error logs can carry a reportable_error metadata field; when Sentry is enabled, a log formatter turns the log message into the event title, uses the wrapped error for cause/stack, attaches optional extra/filepath, and strips reportable_error from console/JSON output.

Adds isReportableError in core so HTTP 200–499 (including statusCode/status on the error cause chain) is treated as non-reportable—used to avoid Sentry noise for provider 4xx, OAuth failures, and similar client errors while still logging them.

Replaces ad-hoc captureCriticalException calls across server, controller, Daytona sandbox, and TrueFoundry dual-write paths with marked logs; removes Daytona's onCriticalAlert hook. Sentry scope now auto-tags tenant/request/session/turn from log context and enables linked errors integration.

Call sites across agent turns, LLM streaming, sandboxes, Redis, request-reply, and API handlers are updated to mark infrastructure/unexpected failures; several paths stop reporting expected cases (missing sandbox versions, authz build failures, duplicate build errors, provider/model client errors). Turn draining adds reporting for unexpected turn.done error/abandoned-timeout cancellations.

^{Reviewed by Cursor Bugbot for commit 3b88b6c. Bugbot is set up for automated code reviews on this repo. Configure here.}

[changeset-bot]

🦋 Changeset detected

Latest commit: 3b88b6c

The changes in this PR will be included in the next version bump.

This PR includes changesets to release 2 packages

Name	Type
@truefoundry/trueforge	Patch
@truefoundry/trueforge-core	Patch

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR

[cursor]

Stale Bugbot comment from a previous run.

[chiragjn]

Keeping it approved with some comments

[cursor]

Stale Bugbot comment from a previous run.

[cursor]

Stale Bugbot comment from a previous run.

[cursor]

Stale Bugbot comment from a previous run.

[cursor]

Stale Bugbot comment from a previous run.

[cursor]

Stale Bugbot comment from a previous run.

[cursor]

Cursor Bugbot has reviewed your changes using default effort 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 c897511. Configure here.}