irs-evidence-research

Pass

Audited by Gen Agent Trust Hub on Oct 4, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data from external official sources including the IRS website and the Electronic Code of Federal Regulations.
  • Ingestion points: The skill searches and reads content from official domains such as irs.gov, uscode.house.gov, and ecfr.gov as defined in SKILL.md and references/source-policy.md.
  • Boundary markers: The instructions mandate a 'citation-gated' output format and include a strict 'Refusal rule' in SKILL.md to prevent unsupported claims or hallucinations.
  • Capability inventory: The skill utilizes a local Python script scripts/validate-citations.py to check for the presence of citations in generated text.
  • Sanitization: Employs specific markdown structures like evidence tables and authority sections to ensure all data is attributed to its source.
  • [DYNAMIC_EXECUTION]: The skill includes a local utility script scripts/validate-citations.py which is intended to be executed via a Python interpreter to validate that the agent's research output contains necessary citations.
  • [EXTERNAL_DOWNLOADS]: The skill references the vendor's GitHub repository (github.com/jupid-tax/jupid-skills) for installation instructions and usage examples. It also facilitates the retrieval of tax guidelines and legal text from well-known official government services.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 4, 2026, 03:20 PM