goal-seeking-agent-pattern
Pass
Audited by Gen Agent Trust Hub on Mar 8, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill references and provides instructions for installing the
amplihackPython package (pip install amplihack) and using its CLI tool. This package is not hosted on a pre-verified trusted domain or organization list. - [COMMAND_EXECUTION]: The provided design patterns and 'Fix-Agent' examples illustrate the use of
subprocess.runto execute various CLI tools (e.g.,ruff,black,pre-commit) based on autonomous decisions made by the agent. - [PROMPT_INJECTION]: The skill documents an architecture vulnerable to indirect prompt injection (Category 8):
- Ingestion points: Agents are designed to process natural language objectives from external files (
--prompt) or inline strings. - Boundary markers: The architecture and examples do not define explicit delimiters or instructions to ignore embedded commands within the 'goal' input.
- Capability inventory: The described agents use tools such as
Read,Grep,Glob, andWebSearch, and they have the capability to execute shell commands viasubprocess.runor theamplihackCLI. - Sanitization: There is no mention of input validation or sanitization of the natural language goals before they are used to generate execution plans.
Audit Metadata