core-dynamic-skills

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

• Third-party content exposure detected (high risk: 0.85). In Inline Mode, the runtime workflow fetches outsider-authored public web content from https://docs.rs/{crate}/... via agent-browser open/get text ".docblock" or WebFetch(...), and then saves/uses that readable documentation to generate skills that are ingested into the agent’s LLM context.

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

• Potentially malicious external URL detected (high risk: 0.90). This skill clearly fetches and ingests runtime content from docs.rs (e.g., "https://docs.rs/{crate}/latest/{crate}/" via agent-browser / WebFetch) and uses that fetched documentation to generate SKILL.md and reference files that directly control the agent's generated prompts/instructions.