The Agent Skills Directory

[PROMPT_INJECTION]: The skill facilitates interaction with untrusted external data by navigating to and snapshotting web pages via the open and snapshot functions. This establishes a surface for indirect prompt injection where malicious instructions embedded in third-party website content could attempt to influence the agent.
Ingestion points: External web content processed in SKILL.md and through the use of interaction templates.
Boundary markers: There are no explicit instructions or markers provided to the agent to treat web content as isolated or potentially untrusted data.
Capability inventory: The skill possesses extensive capabilities including web navigation, element interaction, screenshot capture, and JavaScript execution using the infsh CLI.
Sanitization: No explicit sanitization or filtering of external DOM content is mentioned or implemented in the provided templates.
[COMMAND_EXECUTION]: The skill is designed to run the infsh command-line tool via Bash to manage browser sessions and perform automation. This execution is confined to the specific functions provided by the inference.sh platform.
[EXTERNAL_DOWNLOADS]: Documentation within the skill references setup and installation scripts hosted on the vendor's official GitHub repository at raw.githubusercontent.com/inference-sh/skills. These references are provided for user configuration and setup purposes.

agent-browser