The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to run a "local helper" or "Codex CLI" to render images based on user confirmation. While this is the intended primary purpose of the skill, it involves the execution of local binaries via the shell.
[DATA_EXPOSURE]: The instructions contain a hardcoded absolute path /home/arthur/diagram-prompts/, which leaks specific environmental details (username) associated with the author's local system.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input, such as Mermaid diagrams and descriptions, and interpolates them directly into a generated file template.
Ingestion points: User-provided text descriptions and Mermaid source code are ingested at runtime (referenced in SKILL.md parsing section).
Boundary markers: User content is placed inside markdown blocks within the output file, but there are no explicit safety delimiters or "ignore instructions" warnings to prevent embedded malicious content from affecting the target model when the prompt is later used.
Capability inventory: The skill has file-writing capabilities to the local directory ~/diagram-prompts/ and command execution capabilities for the image generation step.
Sanitization: There is no mention of sanitizing the user-supplied content or the generated file <slug> before writing to the filesystem.

architecture-diagram-gpt-imagegen