packages-update

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

• Third-party content exposure detected (high risk: 0.75). The skill runs scripts/packages.py check/prefetch which shells out to gh release view / gh api .../tags at runtime to fetch upstream GitHub tag/release data (outsider-authored public web content) and then injects the resulting tag/version strings into the agent’s JSON/LLM context.

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

• Potentially malicious external URL detected (high risk: 0.90). The script calls GitHub at runtime (via the gh CLI and nix-prefetch-github, fetching from URLs like https://api.github.com/repos///tags and https://github.com// and then runs builds with nix build), which downloads upstream source and can execute remote build scripts — a runtime external dependency that executes remote code.