tailrocks-swift-agent-integration
Pass
Audited by Gen Agent Trust Hub on Sep 8, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill consists of security-focused instructions that enforce strict boundaries for agent behavior. It explicitly prohibits the model from making unauthorized external actions or disclosing secrets.
- [INDIRECT_PROMPT_INJECTION]: The skill identifies external data as untrusted and provides a defense-in-depth policy to prevent data-driven attacks.
- Ingestion points: Repository files, tool output, fixtures, and registry content (referenced in runtime-trust.md).
- Boundary markers: The skill mandates that embedded instructions in repository files cannot alter authority or side effects, and requires explicit write/command scope to be recorded.
- Capability inventory: The skill utilizes an Xcode bridge for build, test, and preview tasks (referenced in agent-integration.md).
- Sanitization: Instructions require a 'whole-file review' of scripts, hooks, and tool servers before enabling execution, alongside content hash verification for vendored material.
- [DATA_EXFILTRATION]: The skill contains a strong prohibition against credential disclosure, instructing the agent to keep secret values unread and never copy them into logs, prompts, or evidence records.
- [PRIVILEGE_ESCALATION]: The skill enforces a least-capability bridge, requiring manual user consent for Xcode privacy settings and refusing to claims capabilities like UI automation or screenshots that exceed the project context requirements.
Audit Metadata