agent-orchestrator-task

Fail

Audited by Gen Agent Trust Hub on Feb 16, 2026

Risk Level: HIGHPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • Indirect Prompt Injection (HIGH): The skill is designed to ingest 'complex objectives' (untrusted data) and decompose them into subtasks for downstream agents.
  • Ingestion points: Objective descriptions provided via user prompts (e.g., 'Complex Feature Development' examples).
  • Boundary markers: Absent. There are no delimiters or instructions to ignore embedded commands within the processed objectives.
  • Capability inventory: The orchestrator controls 'GitHub Agents' (file-write/version control) and 'SPARC Agents' (methodology execution), providing a high-privilege execution path for malicious instructions hidden in objectives.
  • Sanitization: Absent. No evidence of validation or filtering for the instructions passed to downstream agents.
  • Command Execution (LOW): The pre and post hooks in the YAML frontmatter execute shell commands.
  • Evidence: SKILL.md contains echo, memory_store, and memory_search commands in the lifecycle hooks. While these appear intended for local state management, they demonstrate the capability to execute shell-level operations based on orchestration events.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Feb 16, 2026, 01:32 AM