goal-rules

Pass

Audited by Gen Agent Trust Hub on Nov 5, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The instructions require the agent to autonomously resolve ambiguities by consulting external project resources without human oversight.\n
  • Ingestion points: The agent is instructed to read project documentation, specifications, agent instructions, code conventions, and git history to settle open points (SKILL.md).\n
  • Boundary markers: A 'stop-list' is provided to define forbidden actions (e.g., no main branch pushes, no destructive database operations).\n
  • Capability inventory: The agent has the authority to perform file edits, create branches, commit code, and execute terminal commands.\n
  • Sanitization: There are no instructions for the agent to sanitize or verify the integrity of the guidance found in external project files.\n- [PROMPT_INJECTION]: The skill explicitly commands the agent to ignore standard procedural safety steps such as restating tasks or requesting approval for plans ('Do not restate the task for confirmation, do not offer a plan for approval'). This overrides default agent interaction guardrails to achieve full autonomy.\n- [COMMAND_EXECUTION]: The instructions describe a method to evade 'permission classifiers' by iteratively simplifying blocked terminal commands until they pass, which provides a blueprint for bypassing structural security filters for shell commands.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 5, 2026, 10:39 PM