humanizar
Pass
Audited by Gen Agent Trust Hub on Sep 18, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection as it is designed to ingest and process arbitrary user-supplied text.
- Ingestion points: Untrusted data enters the agent context via the 'texto_fonte' parameter and user re-writing requests.
- Boundary markers: The skill implements a mandatory 'Factual Lock' (TRAVA FACTUAL) that instructs the agent to treat the source text as immutable for facts, and requires a strict triaging process to avoid safety-critical domains.
- Capability inventory: The skill's instructions are limited to text manipulation; there are no subprocess calls, network operations, or file-writing capabilities in the defined scripts.
- Sanitization: The skill uses diagnostic checklists and specific voice profiles to filter and validate content accuracy.
- [SAFE]: The skill operates entirely through instructional logic and markdown guidelines. It does not define any automated scripts, package dependencies, or binaries for execution.
- [SAFE]: The skill references established academic repositories and open-source tools from the NILC/USP research group for linguistic analysis. These external references are documented neutrally for informational purposes.
Audit Metadata