code-and-architecture-improvement
Pass
Audited by Gen Agent Trust Hub on Nov 19, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the repository's documentation and source code to identify tasks and verify changes. This presents a surface for instructions embedded in the codebase to influence the agent.
- Ingestion points: The skill reads
docs/tech/architecture.md,AGENTS.md, and the project's source code files. - Boundary markers: The instructions do not define clear delimiters or warnings to ignore instructions embedded within the files being processed.
- Capability inventory: The skill has the ability to perform git operations (pull, branch, commit, push, merge, delete) and execute local scripts.
- Sanitization: There is no evidence of sanitization or filtering for instructions within the ingested content.
- [COMMAND_EXECUTION]: The skill executes local platform scripts (
scripts/check-*.py) and git commands to manage the refactoring process. These commands run unattended without user confirmation, which increases the impact if the agent is manipulated by indirect injection.
Audit Metadata