The Agent Skills Directory

[SAFE]: Analysis of the skill instructions, Python scripts, and assets confirms that it follows secure practices. It does not perform unauthorized network operations, access sensitive credentials, or use obfuscated logic.
[COMMAND_EXECUTION]: The skill uses local Python scripts, build_diagram.py and validate_diagram.py, to generate and verify HTML artifacts. These scripts are called with structured data and do not provide an arbitrary shell command execution surface.
[PROMPT_INJECTION]: The skill identifies an indirect prompt injection surface through its repository exploration features. This is addressed by the following evidence chain:
Ingestion points: Repository files (docs, entrypoints, schemas) are read during the Explore phase as defined in SKILL.md and references/question-framework.md.
Boundary markers: The agent is instructed to separate facts from inferences and follow a multi-turn questioning framework to close model gaps before planning.
Capability inventory: The skill can write to local files via build_diagram.py, but this is gated by a multi-step approval process and plan confirmation.
Sanitization: The script_safe_json function in scripts/_diagram_utils.py and html_safe_inline_asset in scripts/build_diagram.py provide escaping (e.g., for </ sequences) to prevent structural breakage or script injection within the generated HTML file.

create-diagram