Repository navigation
Labels
enhancementNew feature or requestNew feature or request
Description
Problem or use case
Orca's one-shot AI text helpers each run a single CLI process against one prepared account. The helpers are automatic branch and workspace names, commit messages, PR/MR title and body, Jira titles and independently configured chat-name recipes. When that account's subscription limit is reached, the provider refuses and the request fails. This happens even when the same host has another saved Claude, Codex or OpenCode Go account with capacity left.
- Automatic branch/workspace naming records a failure and backs off. The workspace keeps its placeholder name until a later work event happens after the limit resets, or until the user switches accounts globally.
- Commit-message / PR-field / Jira-title generation shows an error and the user has to switch accounts globally and click again.
Proposed solution
Add an opt-in host setting, off by default, under Settings → AI Provider Accounts. When a helper request is refused for subscription quota, the host that owns the request does the following:
- Waits for the refused process to fully exit, then retries the same request on at most two other eligible saved accounts for the same provider and host. An account is eligible only if it has fresh usage data showing capacity in the relevant bucket.
- Keeps the provider, model, effort, permissions and prompt unchanged, and never touches the global account selection.
- Delivers the result only to the original workspace, draft or session. Stop, a newer request, a manual rename, an owner change or a deleted workspace discards a late result.
- Keeps today's error and backoff when no account is eligible. Account pins, custom commands and credentials, WSL and unsupported routes behave exactly as today.
Alternatives or additional context
- Global preemptive switching ([Feature]: Automatic account failover on rate limit for managed Claude/Codex subscription accounts #20512, feat(claude-accounts): switch accounts before a usage limit #28044, feat(claude-accounts): pane account registry and switch before a usage limit #37567) changes which account everything uses. That is complementary, but it doesn't rescue a request that is already being refused, and it moves unrelated sessions.
- Session-bound chat naming and post-acceptance continuation of editing actions stay with their existing owners (Add opt-in Codex account failover, automatic continuation, and reset warming #21308, Add auto-switch for rate-limited agent accounts #5778, feat(orchestration): managed-account failover and acceptance receipts for supervised workers(受監督工人多帳號遞補與驗收回執) #14290, Native chats follow the account picker and keep the conversation #36649, feat(claude-accounts): move running sessions between accounts through the auth proxy, no restart #37571/feat(claude-chat): native Claude chats are registry rows, follow the pool and move through the proxy #37573).
- Builds on the request-identity / Stop contract from Honor Stop during Source Control AI preparation #28813.
- Related: [Feature]: Automatic account failover on rate limit for managed Claude/Codex subscription accounts #20512, [Bug]: Branch-name generator retries in a tight loop against Claude's session-limit response instead of backing off #9705, Honor Stop during Source Control AI preparation #28813. This issue covers only the one-shot text helpers and should not close [Feature]: Automatic account failover on rate limit for managed Claude/Codex subscription accounts #20512.
Metadata
Metadata
Assignees
Labels
enhancementNew feature or requestNew feature or request
Activity