dlstreamer-coding-agent

CRITICAL E004: Prompt injection detected in skill instructions.

• Potential prompt injection detected (high risk: 1.00). The document contains an explicit, contradictory instruction ("Autonomous execution — never wait for user confirmation") that overrides earlier safety rules requiring explicit user confirmation before running commands, which is a hidden/deceptive behavioral override outside the skill's stated cautious scope.

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 0.90). The skill explicitly requires embedding the verbatim initial user prompt into generated README files, which would force the LLM to reproduce any secrets or API keys the user included in their prompt verbatim in its output.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly resolves repo-root to https://github.com/open-edge-platform/dlstreamer at runtime and contains runtime model-download/export logic that will fetch Hugging Face model repos (e.g. https://huggingface.co/morsetechlab/yolov11-license-plate-detection and https://huggingface.co/PaddlePaddle/PP-OCRv5_server_rec) via hf_hub/optimum-cli (the docs even show --trust-remote-code / trust_remote_code=True), which means remote repository content is fetched during runtime and can execute remote code or control generation/export behavior.