source-research
Pass
Audited by Gen Agent Trust Hub on Sep 4, 2026
Risk Level: SAFEPROMPT_INJECTIONINDIRECT_PROMPT_INJECTIONMETADATA_POISONING
Full Analysis
- [PROMPT_INJECTION]: The skill instructions in SKILL.md explicitly command the agent to "explain how the findings can be used in an offensive capability," which steers the model to generate content potentially associated with malicious use cases.
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the web which could contain malicious prompts.\n
- Ingestion points: External web content retrieved during search operations in quick, deep, and sources modes (SKILL.md).\n
- Boundary markers: None. The skill lacks delimiters to separate user/agent instructions from external data.\n
- Capability inventory: The skill has the ability to browse the web and write research findings to local files (SKILL.md).\n
- Sanitization: None. There is no logic to sanitize or ignore instructions found in external sources.
- [METADATA_POISONING]: The metadata author "GhostWorks" in SKILL.md is inconsistent with the provided author context of "specterops," which can be misleading regarding the skill's origin.
Audit Metadata