generating-flow

Pass

Audited by Gen Agent Trust Hub on Apr 21, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
  • [PROMPT_INJECTION]: The skill uses strong directives to override the agent's default decision-making, requiring strict adherence to a specific 3-step pipeline.
  • [COMMAND_EXECUTION]: The skill instructs the agent to execute a tool in a continuous loop without user intervention or approval checkpoints, which reduces human oversight and increases autonomous behavior.
  • [DATA_EXFILTRATION]: The process involves reading local sfdx project metadata and transmitting it to an external tool. Ingestion point: local sfdx project files via scanning.
  • [PROMPT_INJECTION]: The skill processes untrusted user input and local data to generate executable metadata. Ingestion points: userPrompt and local sfdx project files; Boundary markers: Absent; Capability inventory: execute_metadata_action tool; Sanitization: None.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 21, 2026, 09:48 PM