The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill's architecture relies on the automated installation and execution of external sub-skills. It specifically references repositories such as quangpl/browser-extension-skills and mattpocock/skills in its required-skills.json file. The instructions guide the agent to fetch these dependencies from GitHub at runtime using commit-level pinning (SHAs); however, because these sources are from non-trusted community accounts, they represent a potential remote code execution vector if the source repositories are compromised.
[EXTERNAL_DOWNLOADS]: The skill defines a dependency management system that downloads executable logic and configuration from various GitHub repositories. It utilizes a specialized installer script (install-skill-from-github.py) to pull content from external repositories. While some sources are from well-known organizations (such as Microsoft and Google), others are third-party community repositories that are not pre-verified for security.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection. It takes "plain-language ideas" and requirements from users as input to generate functional browser extension code, project documentation, and task lists. Given its extensive capabilities—including file system access, command execution, and browser automation—a malicious requirement provided as input data could potentially subvert the agent's logic during the development or testing cycle.
Ingestion points: User input for extension ideas and feature requirements defined in SKILL.md.
Boundary markers: None identified; user input is directly used to generate code and tasks.
Capability inventory: File system read/write, shell command execution via sub-skills, and browser control using Playwright MCP (referenced in acceptance.md and playwright-mcp.md).
Sanitization: No explicit sanitization or input validation logic is present in the provided instructions.

browser-extension-launch