macos-local-ipc-integrity
Pass
Audited by Gen Agent Trust Hub on Aug 31, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill consists entirely of instructional markdown and reference documentation. No executable scripts, command-line operations, or package dependencies are present within the files.
- [SAFE]: External references target official Apple Developer documentation and the Model Context Protocol specification, which are trusted resources for platform-specific development.
- [SAFE]: The content explicitly promotes security best practices, such as excluding sensitive data from logs, implementing strict authorization checks, and validating identity and policy after execution suspensions.
- [INDIRECT_PROMPT_INJECTION]: While the skill defines a surface area for IPC data ingestion, it provides specific security mitigations including input size limits, validation of UTF-8/JSON-RPC structure, and the use of authority-owned resource identifiers instead of user-provided paths.
- Ingestion points: SKILL.md and references/framing-reconnect-and-flow.md describe processing arbitrary read chunks from stdio and Unix sockets.
- Boundary markers: The instructions mandate newline-delimited framing for MCP stdio and fixed-width length headers for custom streams to prevent frame confusion.
- Capability inventory: The skill discusses macOS system capabilities (Process, Pipe, NSXPCConnection) in the context of guidance, not as active operations performed by the skill itself.
- Sanitization: The guidelines require rejecting invalid or oversized frames before allocation and redacting sensitive identifiers from diagnostics.
Audit Metadata