The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the Bash tool to execute several commands including node, mkdir, realpath, and which.
The instructions direct the agent to dynamically generate a Node.js script (scripts/export-pdf.mjs) and execute it to perform the conversion.
There is a potential command injection surface in shell command templates like mkdir -p $(dirname <output>), where unvalidated user input for file paths could be exploited depending on the agent's tool-call implementation.
[EXTERNAL_DOWNLOADS]: The skill requires the puppeteer library, which is downloaded from the npm registry.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes external, potentially untrusted HTML content.
Ingestion points: The skill reads local HTML files through the <html-path> argument.
Boundary markers: No boundary markers or instructions to disregard instructions within the ingested HTML are provided.
Capability inventory: The skill has access to Bash (Node.js execution), Write, and Read tools, which could be leveraged if malicious instructions in the HTML are followed by the agent.
Sanitization: The skill does not include steps to sanitize or validate the HTML content before it is rendered by Puppeteer.
Additional Risk: The Puppeteer script launches Chromium with the --no-sandbox flag, which reduces the browser's security boundaries when processing potentially malicious HTML.

export-pdf