human-step-wizard
Pass
Audited by Gen Agent Trust Hub on Nov 8, 2026
Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill follows security best practices for handling sensitive data. It uses interactive hidden prompts (read -s) to capture credentials and restricts secret storage to local environment files with 0600 (user-only) permissions.
- [COMMAND_EXECUTION]: The script wizard_generate.py executes bash -n via a subprocess to validate the syntax of the generated output script. This is a non-interactive, read-only operation used for developer validation.
- [DYNAMIC_EXECUTION]: The generator uses importlib to load its sibling validation module and constructs bash scripts from templates. It ensures safety by single-quoting all specification-derived strings and escaping existing quotes to prevent shell injection.
- [INDIRECT_PROMPT_INJECTION]: The tool processes external JSON specifications to generate executable code.
- [Ingestion points]: The wizard_generate.py script reads a JSON file provided via the --spec command-line argument.
- [Boundary markers]: The generator wraps all specification content in single-quote shell literals in the final bash output.
- [Capability inventory]: The generator writes to the filesystem and runs syntax checks; the resulting wizard executes author-defined shell commands for verification steps.
- [Sanitization]: A dedicated validator script, wizard_spec_validate.py, rejects specifications containing dangerous patterns like eval, source, or unquoted captured variables.
Audit Metadata