egm-review-loop
Pass
Audited by Gen Agent Trust Hub on Nov 3, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [Indirect Prompt Injection Surface]: The skill processes untrusted external data which could potentially contain malicious instructions designed to influence the agent's behavior.\n
- Ingestion points: The skill ingests data from the
TargetFileandGoldfishContext(which can include external documentation URLs), as defined inSKILL.md.\n - Boundary markers: The current instructions do not specify the use of delimiters or 'ignore embedded instructions' warnings for the external content.\n
- Capability inventory: The orchestration logic in
SKILL.mdutilizes file-writing tools to generate reports and file-editing tools to apply repairs to documents.\n - Sanitization: No explicit sanitization or validation of the ingested external content is described before it is passed to the reviewer subagents.\n- [External Content Access]: The skill allows for the retrieval of content from external URLs provided in the
GoldfishContextparameter. This is used to establish a baseline for the automated reviewers.\n- [Automated File Modification]: The skill provides functionality to automatically apply text repairs to files based on findings from the review swarm. While this process includes human-in-the-loop decision gates, the automation of document edits warrants standard review of proposed changes.
Audit Metadata