hypothesis-generation
Pass
Audited by Gen Agent Trust Hub on Oct 1, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from local JSON, CSV, and Markdown files through various validation scripts.\n
- Ingestion points: Scripts such as
validate_hypothesis_schema.py,audit_evidence_ledger.py, andcheck_operationalization.pyingest user-controlled data to perform validation.\n - Boundary markers: The
SKILL.mdinstructions emphasize "Non-negotiable boundaries" and "human accountability," providing context for the agent to treat data as candidate propositions rather than established facts.\n - Capability inventory: The skill utilizes local file read/write operations and command-line execution of bundled Python scripts.\n
- Sanitization: Bundled scripts perform structural, type, and format validation via
scripts/_common.py. Thegenerate_preregistration_scaffold.pyscript applies HTML and Markdown escaping to generated content.\n- [COMMAND_EXECUTION]: The skill instructs the agent to execute bundled Python scripts for scientific data validation and report generation.\n - Evidence: Multiple commands are documented in
SKILL.md, includingpython3 scripts/check_operationalization.pyandpython3 scripts/audit_evidence_ledger.py.\n - Mitigation: The Python scripts are local-only, depend solely on the standard library, and implement strict input validation to prevent common command-line vulnerabilities.
Audit Metadata