The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill documents APIs for reading data from external sources (filesystem, network, and structured formats) and executing system commands, which represents a potential attack surface for indirect prompt injection.
Ingestion points: references/std-fs.md (file I/O), references/std-http.md (HTTP client), and references/std-json.md (JSON parsing).
Boundary markers: Absent. The documentation focuses on language usage and does not explicitly teach LLM-specific data isolation techniques.
Capability inventory: references/std-process.md (std.process.Child for process spawning), references/std-fs.md (file writes), and references/std-net.md (socket operations).
Sanitization: Absent. The skill is a standard programming reference rather than a security hardening guide.
[DYNAMIC_EXECUTION]: The skill provides documentation for loading shared libraries and looking up symbols at runtime, a standard feature of systems programming languages.
Evidence: references/std-dynamic_library.md explains the use of DynLib.open and lib.lookup for interacting with .so, .dylib, or .dll files.
[EXTERNAL_DOWNLOADS]: The documentation contains references to external code repositories for dependency management.
Evidence: examples/build-zig-zon-workflows.md illustrates the use of build.zig.zon with a Git dependency pointing to a public repository on GitHub (github.com/ziglibs/known-folders).
[SAFE]: The skill operates strictly as a documentation reference. The static detector flags for prompt injection in references/builtins.md are false positives triggered by the documentation of standard compiler intrinsics such as @setRuntimeSafety(false), which is presented in an educational context to explain safety-checked vs. unchecked code paths.

zig-0-16