panel-operations

Pass

Audited by Gen Agent Trust Hub on Sep 7, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
  • [SAFE]: The skill serves as an operational manual for a panel-based agent, detailing the use of a predefined set of internal tools (prefixed with panel_) for workflow orchestration.
  • [COMMAND_EXECUTION]: Describes a procedure for crash recovery where the agent is instructed to use a shell to navigate to custom node directories to perform git operations (fetch, pull) and manual source code patching. This is presented as a specific, user-authorized remediation step for addressing technical failures in custom extensions.
  • [EXTERNAL_DOWNLOADS]: Documents the use of the download_model tool to acquire model weights from sources like CivitAI. The instructions explicitly warn against using raw shell commands like curl or wget, favoring the environment's managed download tray for better tracking and safety.
  • [DYNAMIC_EXECUTION]: The crash recovery section involves the agent manually patching source files within custom_nodes directories. This process modifies executable code at runtime to resolve identified software errors.
  • [INDIRECT_PROMPT_INJECTION]: The skill facilitates the loading and analysis of external workflow JSON files, which represents a potential attack surface.
  • Ingestion points: Workflow files are ingested via panel_load_workflow, panel_strip_workflow, and panel_slice_workflow tools.
  • Boundary markers: The provided procedures do not explicitly detail the use of delimiters or specific instructions for the agent to ignore embedded content within the ingested JSON.
  • Capability inventory: The agent possesses the ability to execute nodes (panel_run), modify graph logic (panel_set_widget), and access the shell for troubleshooting/patching.
  • Sanitization: There are no documented steps for sanitizing or validating the contents of the workflow JSON before processing.
  • [SAFE]: The skill contains specific security guidance, advising the agent to avoid using raw shell commands for operations like hardware monitoring (nvidia-smi) or model downloads, steering the interaction toward safer, managed alternatives.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 7, 2026, 01:58 PM
Security Audit — agent-trust-hub — panel-operations