The Agent Skills Directory

[COMMAND_EXECUTION]: The skill requires the agent to execute a local Node.js script (scripts/typeset.mjs) to process Markdown and generate styled HTML. This script is provided within the skill package, has zero external dependencies, and uses built-in Node.js modules for file operations. The execution is limited to formatting tasks using local file paths.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied Markdown content during the typesetting workflow, which presents a surface for indirect prompt injection. This risk is mitigated by explicit instructions and architectural safeguards:
Ingestion points: Untrusted user data is ingested from the source Markdown file during the 'AI Rewriting' and rendering steps.
Boundary markers: The skill provides strict 'Ironclad Rules' in references/formatting-rules.md, instructing the agent not to add facts, delete viewpoints, or alter the core meaning. A quality self-check process is also mandated to verify adherence to these boundaries.
Capability inventory: The agent has the capability to read local Markdown files, execute the provided Node.js rendering script, and write the output HTML file to the user's project directory.
Sanitization: The typeset.mjs script implements several security best practices, including HTML entity escaping, stripping unrecognized HTML tags, and validating URL protocols in links (restricting them to http and https) to prevent cross-site scripting (XSS) or other malicious injections in the final output.

hailaobao-gzh-design