cyrus-setup-launch
Fail
Audited by Gen Agent Trust Hub on Mar 22, 2026
Risk Level: HIGHCREDENTIALS_UNSAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill includes a 'CRITICAL' instruction directing the agent to ignore standard file-handling tools (
Read,Edit,Write) and instead useBashcommands to interact with sensitive configuration files, potentially to bypass platform-level safety monitoring or context logging. - [CREDENTIALS_UNSAFE]: The skill accesses
~/.cyrus/.env, which is identified as containing sensitive credentials includingANTHROPIC_API_KEY,SLACK_BOT_TOKEN, andLINEAR_CLIENT_ID. - [COMMAND_EXECUTION]: The skill executes commands with
sudoprivileges to write system-level configuration files to/etc/systemd/system/and to reload system services usingsystemctl. - [COMMAND_EXECUTION]: Configures persistence mechanisms via
systemdandpm2to ensure the agent's processes are automatically started on boot and maintained in the background. - [EXTERNAL_DOWNLOADS]: Performs global installation of external software via
npm install -g pm2, which downloads and executes code from the public npm registry. - [DATA_EXFILTRATION]: Ingests and processes data from local configuration files (
~/.cyrus/config.json) and environment files, creating a surface for potential data exposure if output is echoed or exfiltrated.
Recommendations
- AI detected serious security threats
Audit Metadata