record-demo
Warn
Audited by Gen Agent Trust Hub on Oct 9, 2026
Risk Level: MEDIUMDYNAMIC_EXECUTIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONCREDENTIALS_UNSAFE
Full Analysis
- [DYNAMIC_EXECUTION]: The camera-operator sub-agent is instructed in
references/camera-operator.mdto write a shell script (.sh) containing automation logic and then execute that script to record the video flow. - [COMMAND_EXECUTION]: The skill executes generated bash scripts and uses the
ffmpegcommand-line tool to perform video to GIF conversions. - [INDIRECT_PROMPT_INJECTION]: The skill interacts with external web applications using
agent-browser. These applications represent an ingestion point for untrusted data that could contain hidden instructions designed to hijack the agent's workflow. No boundary markers or sanitization procedures for this external content are defined in the instructions, and the agent has significant capabilities including file writing and shell execution. - [CREDENTIALS_UNSAFE]: The instructions require the agent to save sensitive browser session data, including cookies and tokens, to a file named
auth.jsonin a temporary system directory to facilitate authenticated flows.
Audit Metadata