gpt56-sol-pro-consult
Pass
Audited by Gen Agent Trust Hub on Jul 25, 2026
Risk Level: SAFE
Full Analysis
- [COMMAND_EXECUTION]: The skill invokes local utility scripts and the
openclitool usingsubprocess.run. These calls use list-based arguments and avoidshell=True, which is a secure implementation practice that prevents command injection. - [DATA_EXFILTRATION]: The skill is designed to transmit local project context to ChatGPT Web for review. To mitigate risk, it includes a mandatory safety scanner (
check_packet_safety.py) that uses heuristic patterns to detect and block the transmission of private keys, AWS access keys, GitHub tokens, and other sensitive credentials. - [INDIRECT_PROMPT_INJECTION]: As the skill ingests local file content for transmission to an external LLM, it possesses an indirect prompt injection surface. The risk is addressed through the use of structured templates (JSON/Markdown), sentinels for output verification, and explicit safety scanning of the final packet.
- [SAFE]: All external references target
chatgpt.com, which is a well-known and expected service for the skill's stated purpose. The logic is transparent, well-tested, and focuses on enhancing model reasoning quality while maintaining credential hygiene.
Audit Metadata