The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill clones the camofox-browser application from github.com/jo-inc/camofox-browser and executes it using node server.js. It also downloads and runs a ~1.3GB Firefox-based browser engine via npm run fetch-bin in scripts/start-camofox.sh.
[EXTERNAL_DOWNLOADS]: The provisioning process fetches application source code, npm packages, and large binary executables from non-whitelisted third-party sources during the start-camofox.sh setup.
[COMMAND_EXECUTION]: The skill uses various shell commands and scripts to manage the environment, including git clone, npm install, uv venv, and nix utilities for library resolution and server management.
[DYNAMIC_EXECUTION]: The generate-closure.sh script dynamically identifies system library paths using nix path-info and ldd, writing them to a file which is subsequently used to set LD_LIBRARY_PATH for the browser execution environment.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest data from bot-hardened websites, introducing a vulnerability surface where malicious site content could attempt to influence the agent's behavior. 1. Ingestion points: Web content retrieved through Camoufox tab sessions. 2. Boundary markers: None explicitly defined in the scripts. 3. Capability inventory: Subprocess execution (node, python, nix), network requests, and workspace file writes. 4. Sanitization: The scripts do not perform specific sanitization or filtering on the scraped web data.

camofox-on-replit