Harden task-pod claim with atomic reservation and rollback

[tomsmith8]

Summary

• Centralize task↔pod claim flow through shared helpers (claimTaskPodAndSetup, claimPodForTaskAtomically, attachPodToTaskAtomically) so reservation, setup, credential persistence, and rollback all go through one path
• Fix race condition where concurrent claims for the same task could each claim a different pod, leaking pods on every retry
• Add rollback: if any post-claim step fails (frontend discovery, control port, credentials), the pod is released back to the pool and task fields are cleared
• Artifact sync path (/api/chat/response) now rejects nonexistent pod links and refuses password-only updates unless the pod row exists
• releasePodById now clears agentUrl alongside podId and agentPassword for symmetrical cleanup
• Capacity read prefers task.podId over usageStatusMarkedBy as the authoritative link
• Remove fake persisted podId: "local-dev" from custom Goose dev paths

Test plan

• Concurrent same-task claim: 5 parallel claims result in exactly 1 pod claimed, 1 task.podId
• Rollback on failure: pod released and task cleared when setup fails after reservation (both /claim-pod and /api/agent)
• Retry safety: failed claim + retry does not consume additional pods
• Attach validation: nonexistent pods rejected, already-assigned pods rejected
• Release symmetry: agentUrl cleared alongside podId and agentPassword
• Artifact attach: valid pod attached via transactional helper; stale pod links rejected
• Phase 2 schema migration (pods.assignedTaskId FK + unique constraint) not yet landed

[vercel]

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

1 Skipped Deployment

Project	Deployment	Actions	Updated (UTC)
hive	Ignored		Mar 25, 2026 3:19pm

[github-actions]

Test environment is now live.

View it at: https://hive-preview-6.sphinx.chat

Database expires at: Mar 25, 2026, 8:21 PM UTC