debug-graph-breaks

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

• Third-party content exposure detected (high risk: 0.85). The skill’s runtime workflow fetches graph-break documentation pages from the public web (e.g., https://meta-pytorch.org/compile-graph-break-site/gb/gbXXXX.html) and reads their text to guide fixes, which is outsider-authored free text injected into the agent’s LLM context.

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

• Potentially malicious external URL detected (high risk: 1.00). The skill explicitly instructs the agent at runtime to fetch graph-break documentation pages from https://meta-pytorch.org/compile-graph-break-site/ (e.g., https://meta-pytorch.org/compile-graph-break-site/gb/gb0059.html), so remote content is retrieved during execution and used to drive diagnostics and fixes.