fix(llm-callsite): pass callSite in approval generators#26508
Conversation
…apper actually triggers
![devin-ai-integration[bot]](https://avatars.githubusercontent.com/in/811515?s=60&v=4){kind=small}
| "approvalCopy", | ||
| "approvalConversation", |
There was a problem hiding this comment.
🚩 New call sites not seeded in latency-optimized migration 040
Migration 040-seed-latency-callsite-defaults.ts:66-74 seeds latency-optimized defaults (fast model, low effort, thinking disabled) for several background call sites including guardianQuestionCopy. The new approvalCopy and approvalConversation call sites are functionally similar (short approval-related LLM calls with tight timeouts of 5s and 8s respectively). Without migration entries, these new call sites fall through to llm.default (opus + max effort), which may be more expensive and slower than intended for what are essentially latency-sensitive background tasks. This isn't a bug per se — the code works correctly — but it may be an oversight given the migration's stated purpose of preventing exactly this cost/latency regression for background tasks.
Was this helpful? React with 👍 or 👎 to provide feedback.
1 participant
Add approval-specific call sites and pass them through sendMessage options so CallSiteRoutingProvider can select per-site routing.
Addresses feedback on #26466.