freeze
Pass
Audited by Gen Agent Trust Hub on Nov 1, 2026
Risk Level: SAFE
Full Analysis
- [COMMAND_EXECUTION]: The skill utilizes local shell scripts to manage state and perform path validation. These scripts are invoked via the
PreToolUsehook to inspect tool inputs (Edit,Write,NotebookEdit) before they are executed. The implementation follows safe practices, such as using--to prevent flag injection in directory traversal and using absolute paths for boundary enforcement. - [DATA_EXPOSURE]: The skill records local usage analytics to a JSONL file in the user's local state directory (
~/.claude/skills/gstack/). This data remains local and is used for tracking skill invocation history. - [INDIRECT_PROMPT_INJECTION]: While the skill ingests a user-provided path via
AskUserQuestion, it applies strict validation. The path is resolved to its physical absolute form usingpwd -Pandreadlink, and it is used solely as a prefix check for future operations. The skill explicitly notes it is not a security boundary against intentional shell command misuse (e.g.,sedviaBash), but rather a safety mechanism to prevent LLM editing errors.
Audit Metadata