deepapi

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

• Third-party content exposure detected (high risk: 0.85). Yes: the required workflow invokes DeepAPI web search/scrape/research endpoints whose responses include scraped snippets/page text from public web and platforms, which the agent then incorporates into the LLM context (classic indirect prompt-injection via retrieved outsider content).

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

• Potentially malicious external URL detected (high risk: 1.00). The skill explicitly auto-updates at runtime by fetching and overwriting its own instructions from https://deepapi.co/deepapi-skill/VERSION.txt and https://deepapi.co/deepapi-skill/SKILL.md, so remote content can directly replace and control the agent's prompts/instructions.

MEDIUM W013: Attempt to modify system services in skill instructions.

• Attempt to modify system services in skill instructions detected (medium risk: 0.40). The skill explicitly instructs the agent to auto-update/overwrite its SKILL.md from a remote URL and to source an env file from the user's home (executing local shell content), which modifies local files and can execute code remotely—but it does not request sudo, system-level file edits, or create user accounts, so the risk is moderate.