us-program-research
Pass
Audited by Gen Agent Trust Hub on Mar 7, 2026
Risk Level: SAFEPROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
- [PROMPT_INJECTION]: The skill exhibits a surface for Indirect Prompt Injection because it processes content from untrusted external sources.
- Ingestion points: In Phase 2 and Phase 3, the agent uses search tools to gather data from student review sites (Reddit, Niche, GMAT Club, GradCafe) and university pages.
- Boundary markers: The prompts provided in
references/subagent-prompts.mddo not include explicit delimiters or instructions for the subagents to ignore potential instructions embedded in the retrieved web content. - Capability inventory: The skill performs file-write operations to the local directory (generating
{CANDIDATE_NAME}_US_PROGRAM_ACTION_PLAN.md). - Sanitization: There is no documented logic to sanitize or filter the external text before it is summarized or included in the final document.
- [DATA_EXFILTRATION]: The skill collects Personally Identifiable Information (PII) during its setup phase.
- Evidence: Phase 0B in
SKILL.mdprompts the user for their 'Full name + email' and detailed academic background. While this is intended for personalizing the resulting research report, it constitutes data collection that users should be aware of.
Audit Metadata