aris-run-experiment

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 1.00). This skill explicitly instructs embedding API keys/secrets verbatim into commands (e.g., wandb login <WANDB_API_KEY>, vastai set api-key YOUR_KEY) and to read them from CLAUDE.md, which requires the LLM to handle/output secret values directly.

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

• Third-party content exposure detected (high risk: 0.80). The skill explicitly calls out-and-integrates with external provider data (e.g., Vast.ai instance listings and CLI outputs referenced in Step 1/Step 2/Step 7 and the /aris-vast-gpu provision flow, plus Modal responses) and uses those third-party responses to choose/rent GPUs and drive deployment, so untrusted third-party content can materially influence actions.