personal-cfo

Warn

Audited by Gen Agent Trust Hub on Jun 16, 2026

Risk Level: MEDIUMEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill instructions in SKILL.md recommend installing the openclaw-personal-cfo package via NPM and cloning a repository from https://github.com/ZhenRobotics/openclaw-personal-cfo.git. These are unverified third-party sources not associated with the established trusted vendor list.
  • [REMOTE_CODE_EXECUTION]: The installation steps for the GitHub repository in SKILL.md include running npm install and npm run build. This allows for the execution of arbitrary scripts defined in the unverified repository's package.json file during the installation and build process.
  • [COMMAND_EXECUTION]: The skill extensively uses shell commands via the cfo CLI alias to manage financial data. User-provided input, such as transaction descriptions and category names, is interpolated directly into these shell commands, which can be seen in the usage examples in SKILL.md.
  • [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface because it processes and stores untrusted user data (transaction descriptions) that is later retrieved and presented to the agent in reports and analysis tasks. * Ingestion points: User-supplied transaction descriptions and categories via the cfo add command in SKILL.md. * Boundary markers: None present; user input is passed as plain string arguments to the CLI. * Capability inventory: Shell command execution for managing the local database and generating reports. * Sanitization: No sanitization or validation of the stored transaction data is mentioned before it is rendered back into the agent's context for analysis.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Jun 16, 2026, 10:03 AM
Security Audit — agent-trust-hub — personal-cfo