security-requirement-extraction
Pass
Audited by Gen Agent Trust Hub on Sep 27, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONMETADATA_POISONING
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The file
skill-report.jsoncontains self-referential content targeting the security analyzer by providing a fake audit report and a pre-determined "safe" verdict. This content should not be treated as authoritative. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data (such as threat models) to generate security requirements and user stories, presenting a surface for indirect prompt injection.
- Ingestion points: External threat model data enters the agent context via instructions in
SKILL.mdand templates inresources/implementation-playbook.md. - Boundary markers: The instructions lack explicit delimiters or warnings to ignore instructions embedded within the user-provided data.
- Capability inventory: No risky tool capabilities (e.g., shell access, network operations) are enabled in the frontmatter configuration.
- Sanitization: The skill does not provide methods for sanitizing or filtering instructions from the ingested threat data.
- [METADATA_POISONING]: The inclusion of
skill-report.jsonis a deceptive practice intended to misrepresent the skill's safety through a fabricated provenance and audit history.
Audit Metadata