The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on executing local shell commands (structurizr.sh, plantuml) and Docker containers (structurizr/structurizr, structurizr/lite) to render and export diagrams. This requires the agent to have permissions to invoke binaries and manage containers on the host system.
[INDIRECT_PROMPT_INJECTION]: The skill has a significant attack surface for indirect prompt injection as it ingests architectural descriptions from user prompts to generate DSL content.
Ingestion points: Architectural descriptions provided by users via natural language prompts are processed in SKILL.md to generate diagrams.
Boundary markers: Absent. The instructions do not specify the use of delimiters or 'ignore embedded instructions' warnings when interpolating user data into the DSL templates.
Capability inventory: The skill utilizes shell command execution, file system writes to /tmp and the current working directory, and Docker container execution.
Sanitization: Absent. There are no instructions for sanitizing user-provided strings before they are used as filenames (e.g., the <name> placeholder in /tmp/<name>.dsl) or as content within the generated DSL, which could lead to command flag injection or DSL escaping.
[DYNAMIC_EXECUTION]: The skill generates domain-specific code (Structurizr DSL or Mermaid) at runtime and immediately executes it using external rendering engines, which is a form of dynamic script generation and execution.

c4-diagram