runpod-deployment

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 1.00). The prompt's example code and HTTP headers embed API keys and bearer tokens directly (e.g., api_key="RUNPOD_API_KEY", Authorization: "Bearer RUNPOD_API_KEY", "YOUR_KEY"), which requires the agent to place secret values verbatim into generated code/requests, creating an exfiltration risk.

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

• Third-party content exposure detected (high risk: 0.90). This skill includes an explicit "Async Handlers" example (reference/serverless-workers.md / SKILL.md) that fetches arbitrary URLs from job["input"] via aiohttp (tasks = [fetch_url(session, url) for url in urls]), meaning untrusted public web content supplied by users can be ingested and directly used in the handler workflow, which could enable indirect prompt-injection or influence runtime actions.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill includes steps that download and execute remote binaries and repository code at runtime — e.g., the GitHub Actions/CI installs and runs runpodctl from https://github.com/runpod/runpodctl/releases/latest/download/runpodctl-linux-amd64 (wget + chmod + run), and model loading/snapshot_downloads pull Hugging Face model repositories (e.g., meta-llama/Llama-3.1-8B-Instruct via the Hugging Face hub) with trust_remote_code=True which can fetch and execute remote Python code, so external content is fetched and executed as a required dependency.