[ianmacartney]

A ConvexError thrown by a workflow step currently reaches the workflow as a plain Error, losing its structured .data. Preserve that data for inline steps and workflow-handler failures, and reconstruct ConvexError when resolving failed steps from the journal. Queued functions and nested workflows carry the same payload through completion callbacks. Historical failures without structured data remain ordinary errors.

Measure the complete serialized failure, including its message/stack and error data, against the existing 800 KiB result budget. Oversized failures become a bounded diagnostic without structured data. Apply the guard before inline/handler failures cross into the component, and when the component processes completion results, so failed steps can be journaled and caught through replay.

Fixes #315. Uses released @convex-dev/workpool 0.4.16, including convex/workpool#257. The peer minimum is ^0.4.16; the changeset requests a minor release for this increased requirement.

Validation:

• Build and all 155 tests, typecheck, and formatting pass. Lint has no errors and three existing filter warnings.
• Native E2E workflows pass on an isolated local backend. Eight additional native error/replay checks pass: all five function execution modes, nested workflows, and oversized inline queries/mutations. Small bigint/byte payloads retain their ConvexError identity and data; oversized failures remain catchable through replay.
• Regression tests cover falsy structured data, exact size boundaries, combined message/data size, oversized messages, and bounded persisted journal failures.

[pkg-pr-new]

Open in StackBlitz

npm i https://pkg.pr.new/@convex-dev/workflow@316

commit: 56ff1f2

[cubic-dev-ai]

All reported issues were addressed across 14 files

<sub>Reply with feedback, questions, or to request a fix.

Re-trigger cubic</sub>