The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the bsk CLI to interact with the Chromium browser, perform navigation, and execute user-defined automation tasks.
[REMOTE_CODE_EXECUTION]: The documentation identifies an auto-update feature where a background daemon automatically upgrades the bsk binary when new releases are published.
[DYNAMIC_EXECUTION]: The tool includes a bsk evaluate command that allows the agent to execute arbitrary JavaScript code within the context of browser tabs.
[DATA_EXFILTRATION]: The skill can access sensitive data such as cookies and active login sessions through its integration with the user's browser. It mitigates this risk through strict 'Red lines' instructions that forbid token theft and credential harvesting.
[INDIRECT_PROMPT_INJECTION]: The skill ingests external content from websites, which could contain instructions aimed at manipulating the agent's behavior.
Ingestion points: Data is gathered via bsk observe, bsk snapshot, and bsk get-html tool calls.
Boundary markers: The instructions do not define specific delimiters for page content, but enforce a bounded workflow focused on success criteria to limit unintended exploration.
Capability inventory: Capabilities include browser navigation, form interaction, and JS execution.
Sanitization: Includes a redaction feature (--redact-values) for trace recordings to prevent the collection of sensitive inputs.

browser-skill