Aias00 opened a new issue, #7043: URL: https://github.com/apache/rocketmq-dashboard/issues/7043
## Current behavior On rocketmq-studio c34535db0 the HTTP engine deliberately sends only a formatting system message and the current user prompt. Prior questions/answers are visible in the Studio transcript but never reach the model, so HTTP follow-up turns have no context. This is a documented current limitation, not a regression. No open implementation PR was found. ## Design and implementation Build HTTP context from a bounded recent slice of this admitted conversation BEFORE the current run start seq. Include only persisted user messages and assistant text, in order; never replay thinking, tool execution, diagnostics or previous system instructions. Use a dedicated role/content message channel, not concatenated role-like prompt text. Cap rows/messages and UTF-8 bytes, keep the current prompt, and expose truncation through an existing notice when history is omitted. Keep legacy single-prompt HTTP and CLI calls unchanged; resume=false explicitly opts out. Enforce conversation isolation, do not fetch all history or add dependencies/schema changes. Retain the formatting system instruction and outbound payload guard. ## Validation HTTP request-capture tests, history builder/service/repository tests, real persistence scoping and bounds, first-turn/opt-out/large-history/Unicode/unreadable-row cases. Update architecture docs to distinguish bounded HTTP context from a CLI resumable session. -- 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]
