untangle-codebase
Pass
Audited by Gen Agent Trust Hub on Nov 5, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process an entire codebase in a single pass. This represents a significant ingestion of untrusted data where malicious instructions could be embedded within the code, comments, or documentation of the analyzed project.
- Ingestion points: The agent is instructed to "Untangle a whole codebase in one pass" and read files to identify patterns like hidden links and special cases.
- Boundary markers: There are no instructions to use delimiters or ignore instructions found within the codebase being refactored.
- Capability inventory: The agent possesses capabilities to read/write files, utilize subagents for task distribution, and execute shell scripts.
- Sanitization: The skill lacks sanitization or validation steps for content extracted from the codebase before it is used to inform the agent's "Redesign from first principles".
- [COMMAND_EXECUTION]: The "Tips" section instructs the agent to generate and run local scripts to assist in the analysis, specifically to "script away the function bodies" and "script one [a dependency graph] that shows what imports what". This requires the agent to produce and execute code based on its parsing of the codebase.
Audit Metadata