The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill references documentation and guidelines from external sources. Evidence: Links to Google's Search Quality Rater Guidelines, Search Central documentation, and SEO guides (references/conceito.md, references/google-search-doc.md). These references are to official and well-known industry resources and do not involve executable code.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources, which is a prerequisite for indirect prompt injection attacks. However, it incorporates significant mitigations. 1. Ingestion points: Text provided by the user and content retrieved from public URLs via the environment's page reading tool (SKILL.md). 2. Boundary markers: Explicit instructions in the 'Procedimento' and 'Regras de governança' sections of SKILL.md require the agent to treat all input strictly as analysis data. The skill also includes a dedicated evaluation case (evals/evals.json, ID 4) to ensure the agent ignores instructions embedded within the audited content. 3. Capability inventory: The skill's functionality is limited to generating a diagnostic report in Markdown format. It does not possess capabilities for file system modification, network exfiltration of sensitive data, or shell command execution. 4. Sanitization: The skill relies on high-level prompt constraints to isolate input content from the agent's control logic.

eeat-audit