The Agent Skills Directory

[Remote Code Execution] (CRITICAL): File references/zvm.md recommends installing ZVM via curl ... | bash from raw.githubusercontent.com/tristanisham/zvm, which executes unverified remote scripts.
[Persistence Mechanisms] (HIGH): File references/zvm.md instructs the agent to modify ~/.bashrc and ~/.zshrc to add ZVM to the system PATH.
[External Downloads] (HIGH): The skill downloads binaries from non-whitelisted repositories tristanisham/zvm and zigtools/zls via curl.
[Indirect Prompt Injection] (HIGH): The skill's ingestion of untrusted project files (build.zig, .zig) combined with its ability to execute build and test commands via MCP tools creates a significant vulnerability surface. Mandatory Evidence: (1) Ingestion points: build.zig, src/*.zig (2) Boundary markers: Absent (3) Capability inventory: zig_build, zig_test MCP tools (4) Sanitization: Absent.

zig