skills/mrvladd-d/memobank/mb-execute/Gen Agent Trust Hub

mb-execute

Warn

Audited by Gen Agent Trust Hub on Mar 7, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands using codex exec and claude CLIs. Specifically, it uses flags like --full-auto and --permission-mode acceptEdits which allow the external tools to modify the filesystem without manual intervention. These commands are constructed via string interpolation of the TASK_ID variable.
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8) because it ingests data from external project files and interpolates it into the execution instructions for sub-agents.
  • Ingestion points: Reads task definitions and requirements from .memory-bank/features/FT-*/..., .memory-bank/requirements.md, and .memory-bank/tasks/backlog.md.
  • Boundary markers: No explicit delimiters or instructions are used to prevent the agent from obeying commands embedded within the task specification files.
  • Capability inventory: The skill has the capability to execute shell commands, create/modify files in the .protocols/ and .tasks/ directories, and spawn sub-agents with write access.
  • Sanitization: There is no evidence of sanitization or validation of the content read from the Markdown files before it is passed to the shell-based execution environment.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Mar 7, 2026, 05:50 AM