tune-mobile-client-performance
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data, which creates a vulnerability surface where malicious instructions could be embedded in data files to influence the agent.
- Ingestion points: The skill processes repository content, profiler traces, logs, recordings, and network captures as defined in
SKILL.mdandreferences/swiftui-swiftdata.md. - Boundary markers:
SKILL.mdcontains a 'Trust boundary' section specifically directing the agent to treat external content as evidence rather than authority and to ignore any embedded commands. - Capability inventory: The skill requires the ability to inspect files, execute build commands, and perform profiling operations to diagnose performance.
- Sanitization: The instructions explicitly require the agent to 'Sanitize private content and identifiers from shared results' and specifically include a test case in
evals/evals.json(ID 6) to ensure the agent refuses to disclose secrets like$SENTRY_AUTH_TOKENfound in data. - [EXTERNAL_DOWNLOADS]: The skill includes a reference to an external source for its diagnostic categories.
- Evidence:
references/swiftui-swiftdata.mdrefers to a specific GitHub repository (github.com/Dimillian/Skills) for its observation and identity categories. - Context: This is a documentation reference to a well-known community repository and does not involve automated code execution or downloading untrusted binaries.
Audit Metadata