jp-law

Pass

Audited by Gen Agent Trust Hub on Sep 15, 2026

Risk Level: SAFEDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [DATA_EXFILTRATION]: The skill performs network operations using curl to interact with laws.e-gov.go.jp. This is the official Japanese government portal for e-Gov Law API V2 and is used to retrieve statutory law information, amendment history, and law text.
  • [INDIRECT_PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection as it ingests and processes text from an external API.
  • Ingestion points: External data enters the agent context via scripts/fetch-law.sh (which retrieves law content) and scripts/search-keyword.sh (which retrieves law snippets).
  • Boundary markers: The skill explicitly instructs the AI to treat script outputs as raw JSON, leveraging the JSON structure as a boundary. SKILL.md also contains explicit behavioral guidelines for the AI to ignore any embedded commands within the retrieved law text.
  • Capability inventory: The skill utilizes curl for network access and standard shell utilities (grep, sed) for text processing. It also reads local definitions from references/law-aliases.md.
  • Sanitization: The skill relies on JSON encoding of the API response and specific AI behavioral prompts to distinguish between data and instructions, providing a robust defensive layer against content-based injection.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 15, 2026, 12:17 PM
Security Audit — agent-trust-hub — jp-law