The Agent Skills Directory

[COMMAND_EXECUTION]: The skill's core functionality involves a JavaScript bridge script (scripts/cavalry-bridge.js) that runs within the Cavalry application, listening on 127.0.0.1:8731 to execute arbitrary JavaScript code received via HTTP POST requests.
[REMOTE_CODE_EXECUTION]: By design, the bridge acts as a local code-execution server. The skill documentation includes a dedicated security section warning that any local process can drive Cavalry through this port without authentication.
[DATA_EXFILTRATION]: The scripting API available through the bridge (api.readFromFile, api.WebClient, api.runProcess) provides capabilities for file system access and network communication, which could be misused to exfiltrate data if a malicious script is executed.
[COMMAND_EXECUTION]: The utility script scripts/cavalry-send.sh uses python3 to parse JSON from the bridge and encode payloads. Automated scanners flagged this as a potential remote code execution pattern, but analysis confirms it is used for local data processing of server responses and command-line arguments.
[SAFE]: The bridge script implements a security check (isCrossOrigin) that rejects requests containing Origin or cross-site Sec-Fetch-Site headers, preventing web-based attacks (CORS/SOP bypass) from the user's browser.
[SAFE]: The bridge advertises its status file location via a GET request and places it in a non-guessable, user-specific preferences directory to prevent local symlink or file-overwrite attacks in shared temporary directories.

cavalry