setup
Pass
Audited by Gen Agent Trust Hub on Sep 4, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill collects user-provided strings for personas, names, and trigger words, which are written into SOUL.md and CLAUDE.md files that influence agent behavior.
- Ingestion points: Phase 6 persona selection and metadata collection.
- Boundary markers: None; user input is written directly to configuration files.
- Capability inventory: The skill uses write_file, read_file, run_code, and list_dir.
- Sanitization: No evidence of sanitization or character filtering for the user-provided inputs.
- [COMMAND_EXECUTION]: The skill executes system-level commands to perform hardware diagnostics and verify installation health.
- Evidence: Use of sysctl, wmic, and uname for RAM and architecture detection in Phase 1, and execution of the betaclaw doctor tool in Phase 7.
- [DYNAMIC_EXECUTION]: The skill uses the run_code tool to dynamically assess environment variables and perform API connectivity tests.
- Evidence: Platform detection via process.platform in Phase 1 and minimal API calls to external AI providers in Phase 7.
Audit Metadata