hai-architecture
Pass
Audited by Gen Agent Trust Hub on Oct 5, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill's primary function is to read and analyze arbitrary source code from a user's repository, which represents a vulnerability to indirect prompt injection if the analyzed code contains malicious instructions.
- Ingestion points: The agent uses tools to read files and search (grep) call chains throughout the target repository to find evidence for its review (documented in
SKILL.mdandSKILL.zh_CN.md). - Boundary markers: The skill attempts to mitigate hallucination and injection through a mandatory 'Evidence Gate' requiring the agent to ground all claims in specific
file:linereferences and reuse the repository's own boundary statements (e.g., arch tests, ADRs). - Capability inventory: The skill utilizes file system reading, grep-based searching, and multi-agent orchestration to analyze code structures and dependency directions.
- Sanitization: There is no explicit sanitization of the code being read; the skill relies on the agent's logical enforcement of its 'Ground Every Claim in Evidence' rule to separate analyzed data from its own instructions.
Audit Metadata