luigidemasi opened a new issue, #21: URL: https://github.com/apache/camel-kit/issues/21
> Imported from [luigidemasi/camel-kit#36](https://github.com/luigidemasi/camel-kit/issues/36), opened by [luigidemasi](https://github.com/luigidemasi) on 2026-05-03. > Closed by commit [cf55a43](https://github.com/luigidemasi/camel-kit/commit/cf55a432eefd922d7f89efb4e7fdb115e3c496be). --- ## Description During testing with `camel-ship`, the LLM selected **Camel 4.18.0 on Quarkus 3.32.0** instead of the correct **Quarkus 3.33.1** from the version-selection guide. The deployed skill file (after Qute substitution) correctly shows 3.33.1, but the LLM ignored the table and used its training data. This is the "two sources of truth" problem from ADR-0025, but at the LLM cognition level rather than the file level — the skill file is correct, the LLM's internal knowledge is wrong. ## Root Cause The HARD-RULE in `version-selection.md` says "Do NOT guess or derive versions" but only applies to downstream skills reading from `config.properties`. It doesn't explicitly tell the LLM "do NOT use your own knowledge for version numbers during version selection itself." ## Fix Strengthen the instructions in `version-selection.md`: 1. Add a HARD-RULE before Step 3 explicitly stating the Quarkus platform BOM version MUST come from the mapping table in Step 2 — not from LLM training data 2. Strengthen the existing HARD-RULE at the bottom to cover version selection itself, not just downstream skills 3. Add explicit "Your training data about Quarkus platform versions is WRONG" phrasing to override LLM confidence ## Files Affected - `camel-brainstorm/guides/version-selection.md` -- 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]
