Claus Ibsen created CAMEL-26875:
-----------------------------------
Summary: camel-jbang - AI tools: the answer to a repeated read
says what to do next
Key: CAMEL-26875
URL: https://issues.apache.org/jira/browse/CAMEL-26875
Project: Camel
Issue Type: Improvement
Components: camel-jbang
Reporter: Claus Ibsen
Assignee: Claus Ibsen
Fix For: 4.24.0
Follow-up from the local-model benchmark (qwen3.6, r32). From the third
identical call of a read tool such as camel_catalog_sample, the tool answers
with a short note: "camel_catalog_sample was already called with these same
arguments 2 times in this session. Its answer does not change, so it is not
sent again: it is in your earlier tool result." The note says what not to do,
but not what to do instead.
In the new set, step runs that asked the same question 4 or more times: r30 13,
r31 5, r32 7, and all 7 failed in r32. In 3 runs of test-citrus-orders step 1
the model asked camel_catalog_sample for "send message", which has no sample,
13 or 14 times in a row ("I apologize for the repeated calls") until its calls
ran out, and never wrote the test.
* The note now says what to do next, and the catalog tools say it in their own
words: a sample or a match that was not there will not be there on the next
call either; go on and write the file with camel_write_file, which checks it
and says what to change.
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)
