The Agent Skills Directory

Unverifiable Dependencies (HIGH): The script validate-markdown.sh utilizes npx -y markdownlint-cli2. This command fetches and executes the package from the npm registry at runtime without version pinning or integrity verification, which is a potential remote code execution vector if the package or registry is compromised.
Indirect Prompt Injection (HIGH): The skill is designed to ingest and process external data (markdown files) and has the capability to modify those files.
Ingestion points: File content from the <filename> argument is read and processed by the linter.
Boundary markers: Absent. There are no instructions or delimiters to prevent the agent from being influenced by instructions embedded within the markdown files being linted.
Capability inventory: The skill executes shell commands and writes to the filesystem using the --fix flag.
Sanitization: None. The content is processed as raw text by the linter tool.
Command Execution (MEDIUM): The skill executes a bash script that triggers npx. While the script path is relative to the skill directory, it allows for arbitrary file path arguments.

markdown-guidelines