research-note
Pass
Audited by Gen Agent Trust Hub on Aug 13, 2026
Risk Level: SAFE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources which could contain malicious instructions.
- Ingestion points: Phase H in SKILL.md performs web searches for recent headlines and analyst sentiment. It also searches SEC filings for qualitative strategic data and risk factors.
- Boundary markers: The instructions do not define specific delimiters or "ignore embedded instructions" warnings for the external data being processed.
- Capability inventory: The skill is designed to save the final report as a local HTML file but lacks high-risk capabilities like arbitrary command execution or network exfiltration of local system files.
- Sanitization: No explicit sanitization, validation, or escaping of the retrieved external text is specified before it is interpolated into the agent context.
- [DATA_EXPOSURE_AND_EXFILTRATION]: No unauthorized data exfiltration patterns were detected. All external data references and citation links are constrained to the vendor's official domain (daloopa.com).
- [METADATA_POISONING]: There is a minor inconsistency where the metadata description mentions Word document generation while the execution logic focuses on HTML reports. This is assessed as a documentation error rather than a deceptive practice.
Audit Metadata