The Agent Skills Directory

[COMMAND_EXECUTION]: The skill requires the use of the belt CLI tool to perform its browser automation tasks. The included templates are Bash scripts that utilize this tool to manage sessions and interactions.
[DYNAMIC_EXECUTION]: The skill provides an execute function that runs arbitrary JavaScript code within the context of the browser session. This is documented for tasks like extracting page titles, cookies, and structured data.
Evidence: references/commands.md and templates/capture-workflow.sh demonstrate the use of execute for running code like document.body.innerText and document.cookie.
[INDIRECT_PROMPT_INJECTION]: As the skill allows an agent to process content from arbitrary URLs, it is vulnerable to instructions embedded in web pages that may attempt to manipulate the agent's behavior.
Ingestion points: Web page content accessed via the open, snapshot, and interact functions, as well as the output from executed JavaScript.
Boundary markers: The instructions do not define specific delimiters or warnings to separate untrusted web content from the agent's internal logic.
Capability inventory: The skill can perform actions like clicking, typing, file uploading, and JavaScript execution, which could be triggered by instructions found in the ingested data.
Sanitization: There is no evidence of filtering or sanitization being applied to the data retrieved from the browser sessions.
[EXTERNAL_DOWNLOADS]: The skill references and provides instructions for installing external components, specifically the belt-sh/cli tool, and links to installation guidelines on GitHub.
Evidence: The installation instructions in SKILL.md point to https://raw.githubusercontent.com/inference-sh/skills/refs/heads/main/cli-install.md.

agent-browser