ratel-decompose-prompt

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

• Third-party content exposure detected (high risk: 0.75). Step 2 explicitly instructs runtime fetching of up-to-date Ratel docs via Context7 MCP and/or WebFetch of docs.ratel.sh (public web content), which can be ingested as readable text 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). Step 2 explicitly instructs fetching runtime docs (WebFetch https://docs.ratel.sh/llms.txt and https://docs.ratel.sh/llms-full.txt, with a GitHub raw fallback https://raw.githubusercontent.com/ratel-ai/ratel/main/README.md) which the skill must pull at runtime to learn the current Skill API and thus directly influences prompt/instruction shaping.