The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The install.sh script fetches and executes the uv installation script directly from a remote source using a shell pipe.\n- [EXTERNAL_DOWNLOADS]: During setup, the skill downloads a large, patched Chromium binary from an external repository to enable its stealth browsing features.\n- [COMMAND_EXECUTION]: The install.sh script utilizes sudo to install a Python execution wrapper in the /usr/local/bin/ directory, which involves modifying system-level directories.\n- [COMMAND_EXECUTION]: The skill provides explicit instructions for the agent to generate and run arbitrary Python scripts for multi-step automation, which involves executing dynamically created code on the host machine.\n- [COMMAND_EXECUTION]: The scripts/eval-js.py utility enables the execution of arbitrary JavaScript expressions within the context of the automated browser.\n- [PROMPT_INJECTION]: The skill is exposed to indirect prompt injection risks by ingesting raw content from third-party websites without defined safety boundaries.\n
Ingestion points: Untrusted content is retrieved via scripts/fetch.py, scripts/screenshot.py, and scripts/pdf.py.\n
Boundary markers: No specific delimiters or instructions to ignore embedded commands are present in SKILL.md.\n
Capability inventory: The skill has the ability to write to system paths (install.sh), execute arbitrary scripts (eval-js.py, Inline Python), and run background network services (scripts/serve.sh).\n
Sanitization: There is no evidence of filtering or sanitizing the data ingested from external pages before it is processed by the agent.

cloakbrowser