The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes a shell command using the pattern npx --yes markdownlint-cli2 "<filepath>". While the file path is wrapped in double quotes, this represents a potential command injection vector if the input path string contains shell metacharacters such as backticks or subshell expansions, especially since no explicit path validation or sanitization is specified in the instructions.
[EXTERNAL_DOWNLOADS]: The skill uses npx --yes to execute markdownlint-cli2, which downloads the package from the NPM registry at runtime if it is not already present on the system. Although markdownlint-cli2 is an established tool for this purpose, the execution of unversioned packages from external registries introduces a dependency on the integrity and availability of the remote source.
[INDIRECT_PROMPT_INJECTION]: The skill reads and processes the entire content of external markdown files provided via the scan context. Because the agent parses this untrusted content to identify accessibility issues, it is susceptible to indirect prompt injection where malicious instructions embedded in a markdown file could attempt to manipulate the agent's logic, findings, or output report.
Ingestion points: Full content of the file specified in the Markdown Scan Context (Step 1).
Boundary markers: None; the agent processes the file content directly without delimiters or instructions to ignore embedded instructions.
Capability inventory: The agent can execute shell commands, read local files, and generate structured textual reports.
Sanitization: No sanitization, escaping, or filtering of the file content is performed prior to the analysis steps.

markdown-scanner