user-file-ops
Pass
Audited by Gen Agent Trust Hub on Jun 23, 2026
Risk Level: SAFEDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [DATA_EXFILTRATION]: The script
scripts/summarize_file.shaccepts arbitrary file paths as input. Without path validation or restricted scoping, this tool could be used to read and expose sensitive files on the host system (e.g., configuration files, private keys) if the agent is directed to summarize them. - [PROMPT_INJECTION]: The skill processes untrusted content from user-provided files or outputs from other skills, creating a surface for indirect prompt injection.
- Ingestion points:
scripts/summarize_file.shreads content from files specified in command arguments (e.g., inSKILL.mdexamples). - Boundary markers: The script's output file format does not include delimiters or instructions to the agent to treat the previewed content as data only.
- Capability inventory: The skill allows reading any accessible file and writing the summary/preview to an output file in the workspace.
- Sanitization: No filtering, escaping, or sanitization of the input file content is performed before inclusion in the preview.
Audit Metadata