The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it analyzes visual data from external web pages that could contain malicious instructions designed to manipulate the agent's behavior.
Ingestion points: Browser screenshots of web pages captured from URLs provided in arguments or inferred from git state.
Boundary markers: No boundary markers or specific 'ignore' instructions are provided to the agent to distinguish between the skill's operational logic and text captured within the screenshots.
Capability inventory: The skill possesses the ability to execute shell commands via agent-browser and Bash to start development servers and interact with the local file system.
Sanitization: Visual content captured from web pages is not sanitized or filtered before analysis by the model.
[EXTERNAL_DOWNLOADS]: The skill utilizes npx to execute the agent-browser tool, which may involve downloading the package from the NPM registry at runtime if it is not already present in the environment.

visual-qa