The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/attach_comment_files.py invokes the gh CLI and npx @playwright/cli using subprocess.run. User-supplied values like repository names and URLs are passed as arguments. Although it avoids shell execution and uses list-based arguments, this represents a potential argument injection surface.
[EXTERNAL_DOWNLOADS]: The skill utilizes npx to run @playwright/cli, a well-known tool from the npm registry. This constitutes an external runtime dependency on a public package registry.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data by reading the content of GitHub comment textareas. This represents a surface for indirect prompt injection.
Ingestion points: The script scripts/attach_comment_files.py reads the value of comment textareas via Playwright's inputValue() method.
Boundary markers: None are present; the content is processed as a raw string.
Capability inventory: File system writes (staging files in a local directory), subprocess execution (gh, npx), and browser-based network access.
Sanitization: Filenames are sanitized using a regular expression to remove special characters before staging; the final data is returned to the agent in structured JSON format.
[DYNAMIC_EXECUTION]: The script generates JavaScript code at runtime by interpolating a JSON payload into a code template and executes it within the browser context using Playwright. This follows a standard template-based execution pattern for browser automation.

shunk031-github-comment-attach-files