transfer-pricing-documentation-master
Pass
Audited by Gen Agent Trust Hub on Jun 16, 2026
Risk Level: SAFEPROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests untrusted data from external sources to generate its documentation reports.
- Ingestion points: The skill reads configuration from
../../CLAUDE.md(related party lists, TP methodology) and processes user-supplied arguments (report year, deadlines). - Boundary markers: Absent; there are no delimiters or instructions to ignore embedded commands within the ingested data.
- Capability inventory: No dangerous capabilities (subprocess, eval, network) were detected in this skill.
- Sanitization: Absent; the external content is interpolated directly into the documentation workflow.
- [DATA_EXFILTRATION]: The skill instructions require reading a file (
../../CLAUDE.md) located outside the skill's own directory structure. While common in some agent environments for loading context, accessing files via directory traversal paths is a data exposure risk pattern.
Audit Metadata