The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill recommends installing software by piping a remote shell script from an unverified GitHub repository (https://github.com/nichochar/install-lightpanda/raw/main/install.sh) directly to the shell: curl -fsSL ... | bash. This practice allows for arbitrary code execution from a third-party source without verification. Found in SKILL.md and references/lightpanda-setup.md.\n- [EXTERNAL_DOWNLOADS]: The skill instructions in SKILL.md and references/lightpanda-setup.md involve cloning source code from an external repository (https://github.com/lightpanda-io/browser.git) for manual builds.\n- [COMMAND_EXECUTION]: The setup documentation in references/lightpanda-setup.md provides instructions for configuring system-level services which require the use of sudo for administrative access on Linux systems.\n- [COMMAND_EXECUTION]: Instructions are provided in references/lightpanda-setup.md to establish persistence on macOS (via launchd) and Linux (via systemd), allowing the browser daemon to persist across sessions.\n- [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface. 1. Ingestion points: open, get text, get html commands (File: SKILL.md, references/commands.md). 2. Boundary markers: Absent. 3. Capability inventory: eval, fill, click, and command-line browser control (File: SKILL.md, references/commands.md). 4. Sanitization: Absent.

lightpanda-browser