paw-mkt-setup
Warn
Audited by Gen Agent Trust Hub on May 20, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [DATA_EXPOSURE]: The scripts 'assets/scripts/chrome-profiles.sh' and 'assets/scripts/chrome-profiles.bat' identify and list file paths to Google Chrome's 'User Data' directory. This location contains highly sensitive data, including browser session cookies, history, and stored credentials.
- [COMMAND_EXECUTION]: The main skill logic in 'SKILL.md' invokes several local Python and shell scripts ('merge-config.py', 'merge-help-csv.py', 'chrome-profiles.sh', 'tool-discovery.sh') to automate project configuration and environment checks.
- [INDIRECT_PROMPT_INJECTION]: The skill processes user configuration and legacy data to generate persistent YAML and CSV files used by other agents. 1. Ingestion points: User input from chat and legacy configuration files in the project root. 2. Boundary markers: None; the data is merged into configuration files without explicit separation from agent instructions. 3. Capability inventory: The skill performs file system modifications (mkdir, mv, rmdir) and updates behavioral configurations ('config.yaml', 'module-help.csv'). 4. Sanitization: Configuration values are handled using standard YAML and CSV writers without filtering for potential prompt injection sequences.
- [DYNAMIC_EXECUTION]: The 'chrome-profiles.sh' script uses 'python3 -c' to execute a Python string at runtime for parsing Chrome's 'Local State' JSON file.
Audit Metadata