omh-ai-slop-cleaner
Pass
Audited by Gen Agent Trust Hub on Nov 9, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONPRIVILEGE_ESCALATION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted source code and project documentation, which constitutes a potential attack surface for indirect prompt injection. Ingestion points include user-supplied files and 'target smell' descriptions provided in SKILL.md. Boundary markers are absent as no specific delimiters or 'ignore' instructions are used to isolate user content from the agent's context. The capability inventory is high, including file system modification for refactoring and shell command execution for tests and linters. No sanitization or filtering of input content is performed before processing.
- [COMMAND_EXECUTION]: The workflow incorporates multiple command-line tools for code analysis and linting, such as ruff, vulture, eslint, knip, and git, as described in references/cleanup-passes.md.
- [DYNAMIC_EXECUTION]: The verification process requires the agent to execute tests and type-checkers on the codebase after modifications to ensure that observable behavior remains identical.
- [PRIVILEGE_ESCALATION]: The skill instructions provide guidance on using the --dangerously-skip-permissions flag when delegating tasks to the Claude Code CLI in non-interactive environments. While SKILL.md recommends using this only in isolated sandboxes or worktrees, providing instructions to bypass standard approval gates is a notable configuration risk.
Audit Metadata