translating-rtl
Pass
Audited by Gen Agent Trust Hub on Sep 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill defines strict operational boundaries, explicitly forbidding the installation of additional skills or tools at runtime. This practice prevents common supply chain and remote code execution attacks. It also restricts network operations to specific technical research handled via the orchestrator, preventing unauthorized data exfiltration.\n- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external data, including product copy, URLs, and user-supplied strings, which represents a potential attack surface. However, the instructions include robust mitigations:\n
- Ingestion points: External project context, core drafts, and product surface content provided at runtime (SKILL.md).\n
- Boundary markers: The instructions mandate the use of semantic isolation mechanisms (e.g.,
<bdi>) and specific Unicode directional controls for embedded LTR runs.\n - Capability inventory: The skill's scope is restricted to text formatting and revision; it lacks direct shell execution or unauthorized network capabilities.\n
- Sanitization: It requires preserving protected values and identifiers exactly as provided and blocks segments if raw values are corrupted, preventing the interpretation of data as instructions.
Audit Metadata