ios-accessibility
Warn
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: MEDIUMMETADATA_POISONINGEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [METADATA_POISONING]: The skill includes deceptive information in its reference documentation, stating that App Store accessibility nutrition labels are required for non-existent versions of Apple software (iOS 26, iPadOS 26, etc.). This misleading information could cause the agent to provide factually incorrect advice regarding platform compliance.\n- [EXTERNAL_DOWNLOADS]: The skill directs the agent and users to an untrusted external domain (sosumi.ai) for technical documentation on AVFoundation characteristics. This domain is not an official Apple resource nor a trusted vendor, posing a risk of exposure to unverified or malicious content.\n- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and audit user-provided SwiftUI and UIKit code but lacks explicit instructions for using boundary markers or sanitizing input. This presents a surface where malicious instructions hidden in code comments or UI data could influence the agent's behavior.\n
- Ingestion points: Audit workflow in SKILL.md and test prompts in evals/evals.json.\n
- Boundary markers: Absent.\n
- Capability inventory: Code auditing and implementation for Apple platforms; no direct shell or network tool access.\n
- Sanitization: Absent.
Audit Metadata