project-health
Fail
Audited by Gen Agent Trust Hub on Mar 13, 2026
Risk Level: HIGHCOMMAND_EXECUTIONCREDENTIALS_UNSAFEDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: Instructions in sub-agent prompts direct the agent to dynamically generate Python scripts, save them to
.claude/scripts/, and execute them for data analysis tasks. - [COMMAND_EXECUTION]: The 'Linux System' permission preset includes
Bash(sudo *), which facilitates execution of commands with root privileges. - [CREDENTIALS_UNSAFE]: Includes permission presets for environment variable passthrough of sensitive API keys (e.g.,
GEMINI_API_KEY,OPENAI_API_KEY) and for thesshpassutility which handles passwords. - [DATA_EXFILTRATION]: Provides presets for network tools (
curl,wget,ssh,scp,rsync) and includes an automated audit task to scan for and report 'leaked secrets' such as API tokens found in project files. - [PROMPT_INJECTION]: The skill processes various local documentation files (
CLAUDE.md,README.md, etc.), creating a surface for indirect prompt injection to influence the agent's behavior during auditing.
Recommendations
- AI detected serious security threats
Audit Metadata