mirror of
https://github.com/discourse/discourse.git
synced 2026-08-06 08:06:54 +08:00
c2a47a6675## Summary Previously, AI bot replies checked agent access against the author of the latest post, which meant a more privileged participant could unintentionally unlock a restricted agent for a conversation; retries could also lose the original authorization context, especially after a failed response. This change keeps topic-selected agents tied to the topic creator’s permissions, records the exact authorizing user before response generation begins, and reuses that information for retries. Explicit agent mentions still use the mentioning user’s permissions, while legacy conversations and missing users fail closed rather than guessing or inheriting access from another participant. ## Source - Patch Triage: https://patch.discourse.org/patch-triage/1180
15 lines
556 B
Ruby
Vendored
15 lines
556 B
Ruby
Vendored
# frozen_string_literal: true
|
|
|
|
module DiscourseAi
|
|
module AiBot
|
|
USER_AGENT = "Discourse AI Bot 1.0 (https://www.discourse.org)"
|
|
TOPIC_AI_BOT_PM_FIELD = "is_ai_bot_pm"
|
|
POST_AI_LLM_NAME_FIELD = "ai_llm_name"
|
|
POST_AI_LLM_MODEL_ID_FIELD = "ai_llm_model_id"
|
|
POST_AI_AGENT_ID_FIELD = "ai_agent_id"
|
|
POST_AI_AGENT_AUTHORIZATION_USER_ID_FIELD = "ai_agent_authorization_user_id"
|
|
TOPIC_AI_AGENT_ID_FIELD = "ai_agent_id"
|
|
TOPIC_AI_AGENT_ID_MAX_LENGTH = 20
|
|
PERSONAL_MESSAGE_CONTEXT = "discourse_ai.ai_bot_personal_message"
|
|
end
|
|
end
|
