fix(sim-cli): report an embedded SimApiError with its own exit code

[waleedlatif1]

Summary

The embedded CLI now reports a SimApiError with that error's own exit code, as the installed CLI already does.

renderEmbeddedError in packages/sim-cli/src/embed.ts returned 1 for every SimApiError and ignored error.exitCode. The terminal renderer (terminal.ts) returns it. So when a workspace operation wait timed out (exit 4, OPERATION_WAIT_TIMEOUT) or the operation needed destination configuration (exit 3, REQUIRES_CONFIGURATION), an in-process run reported a generic failure. The installed CLI reported the real outcome. Now both report the same code. The stderr text is unchanged.

Exit codes this surfaces

• The only SimApiErrors that carry a code other than 1 are the workspace-operation outcomes in workspace-operation-wait.ts:
  • 3 when the operation requires configuration;
  • 4 when the wait times out.
• Wrappers that rethrow an error keep its code.
• Soft exits such as workflows runs wait (0/1/2/3/4) already reached embedded callers through the embed context and are unaffected.

Consumers
Every caller of runEmbeddedCli branches only on zero vs non-zero:

• the Sim sim_cli tool handler;
• agent-CLI sink and curation;
• the Chat worker's result shaping and pipeline stages.

A 3 or 4 that used to arrive as 1 is still a failure to each of them. A caller that wants to tell the outcomes apart, such as a wait that continues after "still running", can now read the real code.

Test plan

• New embed test: an operation wait that times out returns exit 4, the operation receipt on stdout, and OPERATION_WAIT_TIMEOUT on stderr. It failed with exit 1 before the fix and passes after.
• sim-cli suite, bun run lint, type-check (sim-cli and apps/sim), bun run check:audits, and the apps/sim agent-CLI and tool-handler suites.

Follow-ups (not in this PR)

Detached synchronous runs. A workflows run that submits and then polls in bounded slices, so no single request stays open for the whole run. It is not built on async: true, because a queued run is a different run from a synchronous one:

• it gets the async execution timeout instead of the plan's synchronous limit;
• it uses the async rate-limit bucket and queue concurrency;
• a time-based pause keeps polling instead of returning;
• the run resource carries no output for a failed or paused run, while the synchronous response carries the partial output.

It needs:

• a server option to queue a run under the caller's synchronous execution limits and rate bucket;
• the run resource returning partial output for failed and paused runs.

The CLI would then submit with that option, poll with the loop workflows runs wait uses, print exactly what a synchronous run prints when it settles, and on timeout print the run record with exit 4.

Block attribution on the v2 run resource. The polled run resource reclassifies a failed run's stored error string, so it reports no failing block, and a timeout or usage-limit failure known only from its HTTP status is not recognized. Recording the classification (code, plus the block when one is to blame) with the log at error finalization, and applying it on read, would make it match the synchronous response. That change also needs:

• updated public contract descriptions for code, blockId/blockName/blockType, and the run error (runs logged earlier and the resume response keep reclassifying);
• tests that fail when the classification is dropped by log compaction or skipped on the background-execution fallback path.

[vercel]

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

1 Skipped Deployment

Project	Deployment	Actions	Updated
docs	Skipped		Sep 30, 2026 10:32am UTC

[cubic-dev-ai]

No issues found across 2 files

Confidence score: 5/5

• Automated review surfaced no issues in the provided summaries.
• No files require special attention.

_{Re-trigger cubic}

[greptile-apps]

Confidence Score: 5/5

[Medium risk] Changes how the embedded CLI reports error exit codes.

The PR appears safe to merge.

Summary

The embedded CLI now returns a SimApiError’s exit code instead of always returning 1.

• Adds a workspace-operation timeout test that checks exit code 4, the stdout receipt, and the stderr error code.
• Makes the test’s timeout independent of runner scheduling by advancing a mocked clock when the status response is served.

_{Reviews (2) · Last reviewed commit: "test(sim-cli): make the embedded wait-ti..."}

[waleedlatif1]

@greptile

[waleedlatif1]

@cubic-dev-ai review this PR

[cubic-dev-ai]

@cubic-dev-ai review this PR

@waleedlatif1 I have started the AI code review. It will take a few minutes to complete.

[cubic-dev-ai]

No issues found across 2 files

Confidence score: 5/5

• Automated review surfaced no issues in the provided summaries.
• No files require special attention.

_{Re-trigger cubic}