fix(integrations): clear stale task status after connection disconnect (CS-166)

[tofikwest]

Summary

Fixes CS-166 — after a customer disconnected their GitHub integration, related tasks were stuck on "failed" status and the task detail page kept showing historical failed automation runs.

Root cause

ConnectionService.disconnectConnection() soft-deletes the connection (sets status: 'disconnected', clears credentials) but does not touch Task.status or the IntegrationCheckRun audit rows. Three independent gaps result:

• Defect A — stuck task status: when a check from the now-disconnected connection had set Task.status = 'failed', nothing re-evaluates it. The task is pinned to failed indefinitely, because the 12-hour scheduler only processes done tasks.
• Defect B — stale "App Automations" UI: CheckRunRepository.findByTask returned runs regardless of connection status, so the task detail panel still surfaced old failed runs under the active-integrations summary.
• Defect C — scheduler re-marks done→failed: apps/app/src/trigger/tasks/task/task-schedule.ts pulls integrationCheckRuns with no connection filter. For a done task whose review window has elapsed, a stale failed run from the disconnected connection would flip the task to failed on the next tick.

Fix

Minimal, read-side filters + one re-evaluation on disconnect. No schema changes. Check run rows are preserved (matches the existing soft-delete design for IntegrationConnection).

1. ConnectionService.disconnectConnection and deleteConnection now call reevaluateFailedTasksAfterDisconnect(connectionId). For each failed task that had runs from this connection, it re-derives the target status from only active (non-disconnected) automations. Uses a local mirror of the scheduler's getTargetStatus — same 3-way logic (no automations → todo, all passing → done, else failed).
2. CheckRunRepository.findByTask filters out runs whose connection is disconnected.
3. task-schedule.ts adds the same filter on its integrationCheckRuns include.

Test plan

• cd apps/api && npx jest src/integration-platform/services/connection.service → 7/7 new tests pass:
  • Task with no other active automations → todo
  • Task with other passing app automations → done
  • Task with another failing automation → stays failed
  • Latest-run-per-checkId selection works
  • Non-failed tasks untouched
  • No runs from connection → no re-evaluation
  • Failing custom automation keeps task failed
• Full API typecheck: no new errors on changed files
• Manual: connect GitHub, run a failing check, disconnect → task status returns to todo; task detail "App Automations" panel no longer shows old runs
• Manual: with a second active integration passing, disconnect GitHub → task transitions to done

Notes

• Only re-evaluates tasks currently at status: 'failed'. We don't demote done or todo tasks — those transitions already have dedicated flows.
• The 3-way status helper is inlined in ConnectionService rather than extracted. The scheduler's getTargetStatus lives in apps/app and can't be imported from apps/api; a shared package move would be a larger refactor than this bug requires.

🤖 Generated with Claude Code

---

Summary by cubic

Clears stale “failed” task states when an integration (e.g., GitHub) is disconnected and hides runs from disconnected connections so the UI and scheduler ignore outdated failures. Fixes CS-166.

• Bug Fixes
  • On disconnectConnection/deleteConnection, re-derive each affected failed task’s status from active automations only (todo if none, done if all passing; otherwise unchanged). Latest-per-checkId selection is order-independent, and the cleanup is best-effort (wrapped in try/catch) so disconnect/delete still succeed. Audit rows stay.
  • Exclude runs from disconnected connections in CheckRunRepository.findByTask and apps/app/src/trigger/tasks/task/task-schedule.ts so stale failures don’t affect the "App Automations" panel or scheduled status.

^{Written for commit 9f140e1. Summary will update on new commits.}

[vercel]

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

Project	Deployment	Actions	Updated (UTC)
app	Ready	Preview, Comment	Apr 16, 2026 9:04pm
comp-framework-editor	Ready	Preview, Comment	Apr 16, 2026 9:04pm
portal	Ready	Preview, Comment	Apr 16, 2026 9:04pm

[linear]

CS-166 [Bug] - GitHub integration has been disconnected, however still showing failed status

[cubic-dev-ai]

2 issues found across 4 files

Prompt for AI agents (unresolved issues)

Check if these issues are valid — if so, understand the root cause of each and fix them. If appropriate, use sub-agents to investigate and fix each issue separately.


<file name="apps/api/src/integration-platform/services/connection.service.ts">

<violation number="1" location="apps/api/src/integration-platform/services/connection.service.ts:148">
P2: Wrap `reevaluateFailedTasksAfterDisconnect` in a try-catch in both call sites. The connection is already disconnected when re-evaluation runs; if it throws, the primary operation should still succeed rather than surfacing an error to the caller for a best-effort side effect.</violation>

<violation number="2" location="apps/api/src/integration-platform/services/connection.service.ts:255">
P2: The `latestByCheckId` logic drops the `run.createdAt > existing.createdAt` guard present in the original `getTargetStatus`. This makes correctness silently dependent on the query's `orderBy` clause. Add the timestamp comparison to match the original's order-independent behavior.</violation>
</file>

_{Reply with feedback, questions, or to request a fix. Tag @cubic-dev-ai to re-run a review, or fix all with cubic.}

[cubic-dev-ai]

0 issues found across 2 files (changes from recent commits).

_{Requires human review: Auto-approval blocked by 2 unresolved issues from previous reviews.}

[claudfuen]

🎉 This PR is included in version 3.23.2 🎉

The release is available on GitHub release

Your semantic-release bot 📦🚀