cavalry
Fail
Audited by Gen Agent Trust Hub on Aug 3, 2026
Risk Level: HIGHREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The scripts/cavalry-bridge.js file establishes a local HTTP server that listens for incoming POST requests and executes the provided JavaScript code using api.exec() or ui.runFileScript(). This design allows for arbitrary code execution within the Cavalry environment from any local process.
- [COMMAND_EXECUTION]: The scripting API documented in the skill includes api.runProcess() and api.runDetachedProcess(), which allow for the execution of arbitrary system commands and external processes from scripts passed to the bridge.
- [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection. Ingestion points: Data ingested from external JS files and network resources via api.WebClient. Boundary markers: None present. Capability inventory: Execution of arbitrary JS via api.exec() and system commands via api.runProcess(). Sanitization: No sanitization or validation is performed on external inputs before they are incorporated into executable scripts.
Recommendations
- HIGH: Downloads and executes remote code from: http://127.0.0.1:$PORT/get - DO NOT USE without thorough review
Audit Metadata