The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill extracts project metadata such as name and description from local manifest files (e.g., package.json, Cargo.toml, pyproject.toml) and git configuration (SKILL.md, Step 1). This information is interpolated into HTML templates and used in subsequent steps without sanitization, creating a surface where malicious project data could influence the agent's behavior.
Ingestion points: Local project manifest files and git remote output.
Boundary markers: No explicit delimiters or instructions are used to separate untrusted data from the agent's instructions.
Capability inventory: The skill executes shell commands (node -e), performs Git operations (gh api), and writes files to the local system.
Sanitization: The skill lacks validation or escaping for the extracted metadata.
[DYNAMIC_EXECUTION]: The skill guides the user to execute a dynamic Node.js script using the node -e flag to render the social preview HTML to a PNG file using Puppeteer (SKILL.md, Step 4; REFERENCE.md). While the provided script is a template for a specific task, it represents runtime assembly and execution of code.
[COMMAND_EXECUTION]: The skill uses multiple shell commands, including git remote, gh api, and the node interpreter, to perform its core functions of scanning repositories and updating settings.
[EXTERNAL_DOWNLOADS]: The skill references external dependencies and services, including the puppeteer Node.js package and npx @vercel/og (SKILL.md, Step 4). These are sourced from well-known services and trusted organizations.

social-preview