fix(pi): surface provider errors from agent events

[BillLeoutsakosvl346]

Summary

• surface terminal Pi provider errors as failed workflow executions
• ignore transient provider errors while Pi is retrying
• add coverage for provider failures, retries, and successful completion

Type of Change

• Bug fix

Testing

• 54 targeted Pi tests passing
• Sim type-check passing
• lint and repository audits passing
• tested manually with an invalid provider API key

Checklist

• Code follows project style guidelines
• Self-reviewed my changes
• Tests added/updated and passing
• No new warnings introduced
• I confirm that I have read and agree to the terms outlined in the Contributor License Agreement (CLA)

[vercel]

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

1 Skipped Deployment

Project	Deployment	Actions	Updated (UTC)
docs	Skipped		Jul 31, 2026 9:03pm

[cursor]

PR Summary

Medium Risk
Changes how Pi cloud runs fail vs succeed when the CLI exits 0, which affects PR creation and git push gating; logic is localized to event normalization with targeted tests.

Overview
Pi agent_end events now drive workflow failure when the run is actually done: if willRetry is true, the event is ignored so transient provider errors during retries do not abort; otherwise the last assistant message with stopReason error or aborted is normalized to an error event (using errorMessage when present). Successful completions still map to final.

Cloud Pi integration tests were updated to simulate this real stream (including zero CLI exit) and assert the run throws with a scrubbed provider message, stops before commit/push/PR (two sandbox commands only).

^{Reviewed by Cursor Bugbot for commit 9eb6d0c. Bugbot is set up for automated code reviews on this repo. Configure here.}

[BillLeoutsakosvl346]

Demo:

[greptile-apps]

Greptile Summary

Surface terminal Pi provider errors while preserving retry behavior.

• Inspect terminal agent_end events for assistant error and aborted stop reasons.
• Ignore provider errors when Pi explicitly indicates it will retry.
• Add normalization and cloud-backend coverage for terminal failures, retries, and successful completion.

Confidence Score: 5/5

The PR appears safe to merge with terminal provider errors now propagated without treating explicit retries as failures.

The changed event normalization feeds existing error accumulation and propagation paths, while the retry guard prevents transient failures from prematurely terminating execution.

Important Files Changed

Filename	Overview
apps/sim/executor/handlers/pi/events.ts	Correctly distinguishes retrying provider failures from terminal error or aborted agent completion events.
apps/sim/executor/handlers/pi/events.test.ts	Adds focused coverage for terminal failures, aborted runs, retries, and successful agent completion.
apps/sim/executor/handlers/pi/cloud-backend.test.ts	Verifies a zero-exit Pi process with a terminal provider error aborts before repository finalization or PR creation.

_{Reviews (1): Last reviewed commit: "fix(pi): surface provider errors from ag..." | Re-trigger Greptile}

[icecrasher321]

could you explain what this is doing?

[BillLeoutsakosvl346]

Well if willRetry is yes, then the block should't fail, it should try again. A 429 error is an example.