moai-alfred-clone-pattern

Pass

Audited by Gen Agent Trust Hub on Mar 1, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill defines a Master-Clone architecture that is susceptible to indirect prompt injection through its task delegation workflow.
  • Ingestion points: As seen in SKILL.md (Level 2 and 3 architecture) and examples.md, results from initial analysis tasks (like codebase scans) are interpolated directly into the prompt strings of subsequent autonomous implementation tasks.
  • Boundary markers: The prompt templates provided in the documentation do not use delimiters or specific 'ignore' instructions to isolate interpolated data from agent instructions.
  • Capability inventory: Clones are granted high-privilege tool access, including Read, Bash, and Task, which allows them to modify the file system and execute arbitrary shell commands based on the injected instructions.
  • Sanitization: The skill does not include or recommend any validation, filtering, or sanitization of data flowing between autonomous agent instances.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 1, 2026, 01:06 AM