submit
Pass
Audited by Gen Agent Trust Hub on Sep 18, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes a local CLI command to validate the taxpayer's profile.
- Evidence: The skill runs
shinkoku profile --config shinkoku.config.yamlto verify information such as name, address, and My Number registration. - [INDIRECT_PROMPT_INJECTION]: The skill ingests data from local markdown files representing progress from previous steps, which is a potential surface for indirect injection if the project environment is compromised.
- Ingestion points:
SKILL.mdreads from.shinkoku/progress/progress-summary.md,07-income-tax.md,08-consumption-tax.md, and02-assess.md. - Boundary markers: Absent; the skill relies on the structure of its own generated progress files.
- Capability inventory: Local command execution (
shinkokuCLI) and file system read/write operations. - Sanitization: No explicit sanitization of the progress file content is performed before processing.
- [DATA_EXPOSURE]: The skill is designed to handle highly sensitive Personal Identifiable Information (PII), including Japanese tax details and My Number (Social Security equivalent).
- Evidence: The skill instructions specifically guide the agent to verify the presence of My Number and assist in entering tax data into the official e-Tax portal (
https://www.keisan.nta.go.jp/). This is expected behavior for the primary purpose of the skill.
Audit Metadata