The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill performs automated website crawling and content analysis, creating a surface for indirect prompt injection. A malicious website could include hidden instructions designed to manipulate the audit results or exploit the agent's capabilities.
Ingestion points: Website content is retrieved via WebFetch in Phase 1 as described in SKILL.md.
Boundary markers: None present; the agent is instructed to process the raw page data directly for score calculation and categorization.
Capability inventory: The skill utilizes WebFetch for network access, Write for file creation, and Bash for shell access.
Sanitization: No sanitization, escaping, or safety-filtering logic is described for the retrieved web content before it is passed to analysis subagents.
[COMMAND_EXECUTION]: The skill's configuration enables the Bash tool in the allowed-tools section. Although the markdown instructions focus on high-level web analysis and reporting tasks, the availability of a shell environment provides the agent with the capability to execute arbitrary system commands during the audit process.

geo-audit