The Agent Skills Directory

[COMMAND_EXECUTION]: The documentation defines several primitives for shell command execution, including RunCommand, Check, and Service. Specifically, the Service function is described as executing strings through zsh -c to start background dependencies.
[DYNAMIC_EXECUTION]: The skill provides instructions for building and loading Go plugins (shared object .so files) at runtime using go build -buildmode=plugin. This mechanism allows for dynamic extension of the application's executable code.
[INDIRECT_PROMPT_INJECTION]: The skill describes a workflow architecture designed to process external and potentially untrusted data.
Ingestion points: Data enters the agent context through parameters, scoped context, local files, and external research (SKILL.md).
Boundary markers: The guidance suggests using plain English or structured answers but does not require specific delimiters or instructions to ignore embedded commands in the ingested data.
Capability inventory: The workflow framework provides built-in tools for executing arbitrary shell commands and background services (SKILL.md).
Sanitization: There is no mention of explicit sanitization or validation procedures for data before it is used to influence agent decisions or command execution.

gimbal-workflows