agents-docs

CRITICAL E004: Prompt injection detected in skill instructions.

• Potential prompt injection detected (high risk: 1.00). The skill includes a platform-level override ("coordination-exempt: skip the ai-coord gate") that attempts to bypass coordination controls unrelated to its stated purpose of answering product-doc questions, which is a deceptive/out-of-scope instruction.

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

• Potentially malicious external URL detected (high risk: 1.00). The skill runs scripts/fetch-doc.sh at runtime to download external docs (e.g., https://developers.openai.com/codex/codex-manual.md), and those fetched Markdown pages are read into the agent context to drive prompts/instructions.