Baoyuantop commented on code in PR #14006:
URL: https://github.com/apache/apisix/pull/14006#discussion_r4284521678
##########
apisix/plugins/ai-prompt-guard.lua:
##########
@@ -160,7 +160,7 @@ function _M.access(conf, ctx)
messages = new_messages
end
if #messages == 0 then --nothing to check
- return 200
+ return
Review Comment:
Embedding arrays bypass prompt protection
An embeddings request with array input now reaches upstream without
ai-prompt-guard checking its text. get_messages produces no messages for
arrays, so configured deny patterns never run.
##########
t/plugin/ai-prompt-guard.t:
##########
@@ -1038,3 +1041,17 @@ POST /v1/responses
--- response_body
{"message":"Request contains prohibited content"}
--- error_code: 400
+
+=== TEST 45: embeddings array input passes through when there are no messages
to check
+--- request
+POST /hello
+{
+ "model": "text-embedding-3-small",
+ "input": [
+ "hello",
+ "world"
+ ]
+}
+--- error_code: 200
+--- response_body
+hello world
Review Comment:
Test suite exceeds the split threshold
The added coverage extends a 1,040-line test file beyond the roughly
800-line split threshold in test style guidance. Consider moving coverage into
a second test file.
--
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]
