The Agent Skills Directory

[SAFE]: The skill implements secure credential management by instructing the agent to request authentication details from the user at runtime, avoiding the use of hardcoded secrets.\n- [SAFE]: Tool usage is strictly governed by the allowed-tools configuration, limiting command execution to authorized browser automation and basic file system utilities required for report generation.\n- [SAFE]: Session persistence is achieved through local storage of authentication states in a user-controlled output directory, following standard automation practices.\n- [PROMPT_INJECTION]: An indirect prompt injection surface is present due to the ingestion of untrusted web content, though it is mitigated by the skill's restricted capabilities.\n
Ingestion points: The skill retrieves and processes external web content through agent-browser snapshot and agent-browser screenshot commands in SKILL.md.\n
Boundary markers: The instructions do not define explicit delimiters or 'ignore' instructions for the data ingested from web pages.\n
Capability inventory: Capabilities are limited to Bash for browser interaction (agent-browser) and basic file management (mkdir, cp), which are appropriate for the tool's documented purpose.\n
Sanitization: No automated sanitization or filtering of external web content is performed.

dogfood