davsclaus opened a new pull request, #27422: URL: https://github.com/apache/camel/pull/27422
https://issues.apache.org/jira/browse/CAMEL-25371 In the r5 run of the local-model benchmark, three of the steps that used all 20 tool calls called `camel_validate_source` with byte-identical content 16, 10 and 10 times. Each time the answer came back in full. The note for repeated calls, sent from the third identical call on, only covers tools marked deterministic. `camel_validate_source` is not marked, because without `content` it reads the file on disk, which changes. ## Changes - **`ToolDescriptor.deterministicWhen(param)`**: a tool is deterministic for the calls that pass that argument. `camel_validate_source` uses it with `content`; validating the file on disk still runs every time. - **`ToolDescriptor.repeatHint(text)`**: what the repeated-call note adds for that tool. For validation it reads: change the line the error names, then validate the changed content, or write it with `camel_write_file`. - **`camel_validate_source`**: when the given content is the file on disk unchanged, the answer has `sameAsFile: true` and the note "This content is the same as <file> on disk: an edit you meant to make is not in it." One of the stuck steps validated the old version 9 times. The camel-jbang-mcp server and the AI panel both pass the call's arguments to `RepeatedToolCalls.repeatOf`, so they pick this up as it is. Tests: new cases in `RepeatedToolCallsTest` and `AuthoringToolsTest`; the camel-jbang-core suite passes (1512). 🤖 Generated with [Claude Code](https://claude.com/claude-code) https://claude.ai/code/session_01STT6whBgK1AqsSsUKrnE8m -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
