The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to fetch and analyze content from external, user-supplied URLs to perform a quality assurance audit.
Ingestion points: Untrusted data enters the agent context through the fetching of HTML, robots.txt, and sitemaps from target websites as defined in SKILL.md (Step 2: Raccolta prove).
Boundary markers: The instructions lack specific technical delimiters to wrap external content, although they include strong behavioral instructions ('Regola d'oro') and 'Honesty Gates' to ensure the agent only reports verified facts.
Capability inventory: The agent is authorized to use shell tools (curl), fetch web content, and potentially use an agentic browser, while also generating report files.
Sanitization: There are no instructions for sanitizing or escaping the fetched web content before it is processed by the agent, allowing a malicious site to potentially embed instructions aimed at biasing the audit results.
[COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands to gather technical data.
Evidence: SKILL.md explicitly directs the agent to use curl -sI https://<dominio> to verify HTTP headers and redirect chains.
Evidence: reference/checklist.md suggests the use of diagnostic CLI tools such as npm audit, composer audit, and security scanners (WPScan, Snyk) on the target environment.

checklist-sito-web