The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests and processes data from external, untrusted sources (websites and PDFs).
Ingestion points: lib/crawler.js (retrieves web content via Playwright) and lib/pdf-reader.js (downloads and parses PDF text).
Boundary markers: Absent. While content is structured into Markdown files in the docs/ directory, there are no explicit delimiters or instructions telling the AI to ignore embedded commands within that content.
Capability inventory: The skill can write files (fs.writeFileSync, fs.cpSync), delete files (fs.rmSync), and execute shell commands (child_process.execFileSync to call the Playwright tool).
Sanitization: Content extracted from documents is not sanitized for natural language instructions that might influence the Agent's behavior during the skill generation or analysis phase.
[EXTERNAL_DOWNLOADS]: The skill downloads external PDF files from user-provided URLs using the lib/pdf-reader.js module. It uses the standard http and https modules to fetch these files into a temporary directory for processing.
[COMMAND_EXECUTION]: The skill executes local commands to interact with the Playwright tool and perform file system operations.
Subprocess execution: lib/crawler.js uses execFileSync to invoke the playwright skill's tool.js for web scraping.
File system operations: tool.js and lib/generator.js perform significant file system operations including creating directories, writing generated code and documentation, and copying or deleting files in the installation target path.

doc-skill-generator