agency-minimal-change-engineer
Pass
Audited by Gen Agent Trust Hub on Oct 5, 2026
Risk Level: SAFENO_CODEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill contains no executable scripts, shell commands, or network operations. It is composed of natural language instructions and illustrative code snippets that do not trigger automated actions.- [NO_CODE]: The repository consists solely of instructional markdown content designed to influence the behavior of an AI agent. It does not include dependencies or scripts that could be exploited for execution.- [INDIRECT_PROMPT_INJECTION]: The skill defines an agent that processes untrusted external data (source code) and possesses file-writing capabilities. While it does not contain malicious code, it operates within an environment where data could influence behavior.
- Ingestion points: The agent reads source code files and task descriptions (SKILL.md).
- Boundary markers: No explicit prompt boundary markers are defined to isolate task instructions from code content.
- Capability inventory: The agent uses file-read and file-write tools to perform engineering tasks.
- Sanitization: No specific input sanitization or validation logic is provided for the code files it processes.
- Mitigation: The skill's primary identity as a "Minimal Change Engineer" acts as a behavioral guardrail by strictly limiting the agent to the smallest possible modifications, thereby reducing the blast radius of any indirect injection attempt.
Audit Metadata