full-council
Pass
Audited by Gen Agent Trust Hub on Nov 7, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted external content, including code files, git diffs, and pull request data, which could contain embedded malicious instructions aimed at overriding agent behavior.
- Ingestion points: Untrusted data enters the context in
SKILL.mdduring the identification phase where the orchestrator collects repository facts, diffs, and code context to provide to the sub-agents. - Boundary markers: There are no explicit boundary markers or XML-style delimiters defined in the prompt to separate the untrusted code content from the agent's instructions.
- Capability inventory: The skill has permissions to read files and delegate tasks to other models via
delegate_task. All components are explicitly restricted to read-only operations, preventing file modifications, commits, or external network exfiltration. - Sanitization: The instructions do not specify any sanitization, escaping, or filtering of the code or diff content before it is interpolated into the briefs for the Analyst agents.
Audit Metadata