Handle 404 in waitForTaskActive when task is deleted by concurrent run

[blinkagent]

Problem

When a synchronize (push) event triggers the doc-check workflow for a PR that already has an existing task, the action finds the task via the list endpoint and enters waitForTaskActive to poll until it's active + idle. However, if a concurrent workflow run (from a previous push) is still executing its cleanup phase, it can delete the task between the initial list call and a subsequent getTaskById poll.

This results in an unhandled 404 error that crashes the action:

Coder API error: Not Found
statusCode: 404
response: '{"message":"Resource not found or you do not have access to this resource"}'

Stack trace:

at RealCoderClient.request (dist/index.js:26727:13)
at async RealCoderClient.getTaskById (dist/index.js:26780:22)
at async RealCoderClient.waitForTaskActive (dist/index.js:26802:20)
at async CoderTaskAction.run (dist/index.js:27001:7)

Example failure: https://github.com/coder/coder/actions/runs/23586039175/job/68679282993#step:7:65

Root Cause

waitForTaskActive polls getTaskById in a loop but does not handle the case where the task disappears (404) during polling. This is a race condition between concurrent workflow runs for the same PR:

1. Run A finds existing task doc-check-{N}, enters waitForTaskActive
2. Run B (previous run) reaches its always() cleanup step and deletes doc-check-{N}
3. Run A's next getTaskById poll returns 404, action crashes

Suggested Fix

In waitForTaskActive, catch 404 errors from getTaskById and either:

• Treat it as a signal to re-create the task (fall through to the create path)
• Retry with a fresh task lookup via the list endpoint
• Return a specific error that the caller (action.ts) can handle to create a new task

Additionally, consider using GitHub Actions' concurrency groups to prevent overlapping runs for the same PR, which would avoid the race condition entirely.

---

Created on behalf of @johnstcn