The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs users and agents to install the Steel CLI using the command curl -sSf https://setup.steel.dev/install.sh | sh. This is a high-risk pattern that executes remote code from the internet directly in the shell without any verification of the script's contents or integrity.
[COMMAND_EXECUTION]: The skill relies on the steel command-line tool for all operations. This includes starting browser sessions, executing JavaScript, and managing persistent session states on the local system.
[EXTERNAL_DOWNLOADS]: The skill requires downloading the CLI tool and suggests installing various SDKs (steel-sdk, playwright, puppeteer) from public registries (NPM/PyPI).
[DATA_EXFILTRATION]: The steel browser cookies and steel browser storage commands provide access to sensitive session data. If the agent is compromised via indirect injection, these tools could be used to harvest and exfiltrate user credentials or session tokens.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection due to its core function of processing arbitrary web content.
Ingestion points: steel scrape and steel browser navigate fetch content from untrusted external URLs.
Boundary markers: The skill does not provide instructions to the agent to distinguish between its own operational instructions and content found on target websites.
Capability inventory: The toolset includes steel browser eval, which allows execution of arbitrary JavaScript in the browser context, and the ability to capture screenshots or PDFs.
Sanitization: There is no evidence of sanitization of the scraped content before it is provided to the agent.

steel-browser