swiftui-core
Pass
Audited by Gen Agent Trust Hub on Oct 13, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONMETADATA_POISONING
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill defines a mandatory workflow that requires spawning a research agent (
fuse-ai-pilot:research-expert) which uses external search tools (Context7/Exa) to verify documentation. This creates an ingestion surface where untrusted data from the web can enter the agent's context. - Ingestion points:
SKILL.md(Agent Workflow section). - Boundary markers: Absent; no instructions are provided to delimit or treat the research findings as potentially untrusted content.
- Capability inventory: The skill is scoped to SwiftUI development and UI verification; while it orchestrates other agents, the skill itself does not contain direct file-system write, network exfiltration, or arbitrary command execution capabilities.
- Sanitization: Absent; there are no requirements for validating or filtering the retrieved documentation.
- [METADATA_POISONING]: The skill's YAML frontmatter and instructional content reference non-existent or speculative versions of Apple software (e.g., iOS 26, iOS 27, SwiftUI 6) and hardware (iPhone Duo). This deceptive metadata can mislead the agent into attempting to use hallucinated APIs (e.g.,
.glassEffect(),ArrangementView) that do not exist in the current development environment.
Audit Metadata