The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill directs the installation of the browserman-cli package globally via npm install -g browserman-cli@latest. This package is the primary interface for the service and is associated with the skill's vendor infrastructure.
[REMOTE_CODE_EXECUTION]: The skill utilizes npx -y browserman-cli@latest to dynamically download and execute code from the npm registry. This pattern involves executing remote code at runtime, although it targets the vendor's official package.
[COMMAND_EXECUTION]: The agent executes various subcommands via the browserman CLI, such as browser ping, page open, and page screenshot, to automate browser interactions on the user's behalf.
[DYNAMIC_EXECUTION]: The skill supports running local JavaScript scripts using browserman script run --local. These scripts are executed with local environment permissions. The instructions include a specific warning for the agent to review the code for trustworthiness before execution.
[INDIRECT_PROMPT_INJECTION]: The skill creates a vulnerability surface by reading arbitrary web content using browserman page read. Malicious instructions embedded in processed pages could attempt to influence the agent's behavior.
Ingestion points: Data retrieved from external URLs via browser interaction commands in SKILL.md.
Boundary markers: None provided to isolate untrusted web data from system instructions, though a 'destructive-action checklist' requires user confirmation for high-risk operations.
Capability inventory: Subprocess execution, local filesystem access (screenshots/scripts), and full browser control.
Sanitization: No explicit sanitization or filtering of ingested web content is documented.
[CREDENTIALS_UNSAFE]: The agent is instructed to read sensitive authentication tokens and browser identifiers from the local configuration file at ~/.browserman/config.json. This access exposes local session credentials to the agent's processing context.

browserman