mirror of
https://github.com/github/codeql.git
synced 2026-06-19 11:51:08 +02:00
8e5f214041
Verified all prompt-injection framework models against the real Python SDK sources: - OpenRouter: the official openrouter SDK uses client.chat.send(messages=) (not chat.completions.create), client.embeddings.generate(input=) (not embeddings.create), and client.responses.send(input=, instructions=). Corrected the framework qll and model, and fixed the test files that used the wrong API. - Anthropic: added the managed-agents system prompt sink (beta.agents.create/update Argument[system:]). - Google GenAI: added models.edit_image Argument[prompt:] as user content. OpenAI, agents and LangChain models were confirmed correct against their SDK sources. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
17 lines
688 B
YAML
17 lines
688 B
YAML
extensions:
|
|
- addsTo:
|
|
pack: codeql/python-all
|
|
extensible: sinkModel
|
|
data:
|
|
# `responses.send` instructions is a system-level prompt; input is user content
|
|
- ['OpenRouter', 'Member[responses].Member[send].Argument[instructions:]', 'system-prompt-injection']
|
|
- ['OpenRouter', 'Member[responses].Member[send].Argument[input:]', 'user-prompt-injection']
|
|
# Embeddings input is user-level content
|
|
- ['OpenRouter', 'Member[embeddings].Member[generate].Argument[input:]', 'user-prompt-injection']
|
|
|
|
- addsTo:
|
|
pack: codeql/python-all
|
|
extensible: typeModel
|
|
data:
|
|
- ['OpenRouter', 'openrouter', 'Member[OpenRouter].ReturnValue']
|