The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted markdown files to extract Mermaid diagrams. This represents a vulnerability surface where maliciously crafted markdown content (diagrams or headers) could attempt to influence the agent's behavior.
Ingestion points: The scripts/extract_diagrams.py script reads the content of a user-specified markdown file.
Boundary markers: No explicit delimiters or instructions are provided to the agent to ignore potentially malicious content within the markdown file.
Capability inventory: The skill executes shell commands, writes files to the disk, and launches a headless browser via mermaid-cli.
Sanitization: Filenames generated from headers are sanitized to be alphanumeric, but the content of the Mermaid diagrams is passed directly to the renderer.
[COMMAND_EXECUTION]: The skill uses scripts/extract-and-generate.sh to orchestrate file operations and external tool execution.
The script executes mmdc (Mermaid CLI) and Python scripts with arguments derived from user input (file paths).
It uses mkdir -p to create directories based on provided output paths.
[EXTERNAL_DOWNLOADS]: The documentation provides instructions for the user to download and install system dependencies from well-known and trusted sources.
Fetches the Google Linux signing key from https://dl.google.com/linux/linux_signing_key.pub to facilitate Chrome installation.
References the official Google Chrome repository for Debian-based systems.

mermaid-tools