harbor-adapter-creator

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.90). The skill's adapters explicitly load data from open public sources (e.g., SimpleQA's pandas.read_csv(DATA_URL) fetching an OpenAI public CSV, GAIA/HuggingFace attachments, CodePDE's HuggingFace downloads in test.sh and repo clones) and then populate instruction.md and task files that agents will read and act upon, so untrusted, user-controlled third-party content can directly influence agent behavior.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). The SimpleQA adapter loads data at runtime via pandas.read_csv("https://openaipublic.blob.core.windows.net/simple-evals/simple_qa_test_set.csv"), and those CSV records are directly used to generate instruction.md/ground_truth (i.e., the prompts the agent sees), so this URL is a runtime external dependency that directly controls agent prompts.