rust-refactor-helper
Fail
Audited by Gen Agent Trust Hub on Oct 10, 2026
Risk Level: HIGHOBFUSCATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [OBFUSCATION]: The skill instructions contain hidden Zero-Width Space characters (U+200B) strategically placed before markdown code block markers. These characters are invisible to users but present in the file data, which is a technique used for obfuscation or to bypass automated content filtering.
- Evidence: Found in sections 'Selected Code', 'Extracted Function', 'Replacement', and 'New File Structure' preceding the triple backtick delimiters.
- [INDIRECT_PROMPT_INJECTION]: The skill architecture is designed to ingest, analyze, and modify source code from the agent's environment, creating a vulnerability surface where instructions embedded in project files could hijack the agent's logic.
- Ingestion points: Rust source code files (.rs) and project documentation (.md) read via the 'Read', 'LSP', 'Grep', and 'Glob' tools.
- Boundary markers: The skill instructions do not define any specific delimiters or 'ignore' commands to help the agent distinguish between code-to-be-refactored and instructions embedded within that code.
- Capability inventory: The skill has the 'Edit' tool enabled, which allows the agent to commit changes to the filesystem based on potentially poisoned input.
- Sanitization: There is no evidence of validation or sanitization of ingested content before it is processed by the refactoring logic.
Recommendations
- AI detected serious security threats
Audit Metadata