PromptInjection

Warn

Audited by Gen Agent Trust Hub on Mar 5, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [COMMAND_EXECUTION]: The skill mandates the execution of a curl command to http://localhost:8888/notify immediately upon invocation as part of a required voice notification step.
  • [COMMAND_EXECUTION]: The reconnaissance workflow facilitates the generation and execution of a local shell script (reconnaissance.sh) that performs file system operations and system calls to manage browser tools.
  • [PROMPT_INJECTION]: The skill definition employs mandatory behavioral overrides to force the agent into an offensive security role and execute background network tasks.
  • [EXTERNAL_DOWNLOADS]: The documentation references well-known security packages from public registries such as promptfoo, garak, and pyrit for automated testing.
  • [PROMPT_INJECTION]: High risk of indirect prompt injection during reconnaissance. 1. Ingestion points: browser extraction and network logs identified in APPLICATION-RECONNAISSANCE-METHODOLOGY.md. 2. Boundary markers: None identified in the processing logic. 3. Capability inventory: subprocess execution (bash), local network access (curl), and file system write access. 4. Sanitization: Absent from the data ingestion flow.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Mar 5, 2026, 07:38 AM