The Agent Skills Directory

[Command Execution]: The skill uses local shell commands (mkdir, cp) and a browser automation tool (agent-browser) to organize reports and execute test steps. This activity is focused on the local output directory and the specified target application.\n- [External Content Interaction]: By design, the skill processes information from external web pages, which is a potential vector for indirect prompt injection. (1) Ingestion points: Web content is ingested from the user-defined target URL as described in SKILL.md. (2) Boundary markers: Explicit delimiters for browser-retrieved content are not defined. (3) Capability inventory: Includes file system access (mkdir, cp) and browser interaction via agent-browser. (4) Sanitization: The skill does not explicitly sanitize the external web content. This risk is inherent to web-automation tasks and is mitigated by the skill's specific testing logic.\n- [Session Data Management]: The skill provides functionality to save authentication states to a local file (auth-state.json). While necessary for testing authenticated workflows, users should treat the contents of the output directory as sensitive.

dogfood