affiliate-payout-audit
Pass
Audited by Gen Agent Trust Hub on Oct 26, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONNO_CODE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external affiliate data (CSV/spreadsheets) and instructs the agent to write and execute scripts for math recomputation. This creates a surface for indirect prompt injection where malicious instructions embedded in input data could influence agent behavior.
- Ingestion points: Affiliate platform exports and internal billing/order extracts (SKILL.md, Step 2).
- Boundary markers: Absent; the skill provides no guidance on using delimiters or ignoring instructions within the ingested data.
- Capability inventory: The skill explicitly recommends the agent write and execute scripts for full-population recomputation (SKILL.md, Step 10).
- Sanitization: Absent; the instructions do not include steps to sanitize or validate data fields before processing.
- [NO_CODE]: The skill consists entirely of instructional and reference documentation in Markdown and JSON formats, with no executable code files (.py, .js, .sh) or binary components included in the package.
Audit Metadata