ai-bot-builder
Pass
Audited by Gen Agent Trust Hub on Oct 6, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONPROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill architecture handles untrusted user input and external data (e.g., KB articles). This surface is mitigated by instructions to use XML delimiters (<user_input>) and explicit 'ignore instructions' directives in system prompts, as detailed in references/injection-and-jailbreak-defense.md.- [COMMAND_EXECUTION]: The script scripts/test_bot_scripts.py uses subprocess.run to perform unit testing on the skill's other utility scripts. This is standard developer tooling and is not exposed to user-controlled inputs in a live agent context.- [PROMPT_INJECTION]: Several files contain patterns matching instruction overrides (e.g., 'Ignore previous instructions'). These are explicitly provided as examples of attack vectors within a defensive context to teach the agent and developers how to recognize and block such attempts.- [DYNAMIC_EXECUTION]: The testing script scripts/test_bot_scripts.py uses importlib to dynamically load and test sibling scripts. This uses static local paths and follows safe development practices for modular testing.
Audit Metadata