agent-batch-guard
Pass
Audited by Gen Agent Trust Hub on Mar 10, 2026
Risk Level: SAFECOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The provided Python template in SKILL.md uses subprocess.run to execute adb (Android Debug Bridge) commands, allowing the agent to dump UI hierarchies and pull files from connected mobile devices.
- [REMOTE_CODE_EXECUTION]: The skill promotes a Scripted Batch Processing pattern where the AI agent is instructed to write Python scripts to local storage and execute them via the shell, representing a dynamic code generation and execution risk.
- [PROMPT_INJECTION]: The skill provides behavioral instructions for the AGENTS.md configuration using high-priority directives like 黄金规则 (Golden Rules) to override standard agent looping logic. 1. Ingestion points: Data is ingested from external Android devices via adb shell uiautomator dump as described in SKILL.md. 2. Boundary markers (absent): The recommended script templates do not include delimiters or instructions to ignore potential commands embedded in the scraped UI XML data. 3. Capability inventory: The skill facilitates shell command execution (adb), file system writes, and the execution of generated scripts in SKILL.md. 4. Sanitization (absent): There is no evidence of input validation or escaping for the data retrieved from external devices before it is processed or archived.
Audit Metadata