multi-ai-collab

Pass

Audited by Gen Agent Trust Hub on May 12, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill coordinates sub-agents by executing shell commands to invoke external CLI tools like Codex, Gemini CLI, and Claude Code. It utilizes command substitution (e.g., $(cat src/file.ts)) to pass the content of local files as input to these tools.
  • [EXTERNAL_DOWNLOADS]: Documentation within the skill recommends the installation of official CLI utilities from well-known and trusted providers, including Google's @google/gemini-cli and Anthropic's Claude Code. These references target established, official package registries and repositories.
  • [PROMPT_INJECTION]: The skill exhibits an attack surface for Indirect Prompt Injection, common in orchestrator patterns.
  • Ingestion points: Local file content is read and interpolated directly into prompts via shell command substitution in multiple workflow templates (e.g., SKILL.md, references/workflows/pipeline.md).
  • Boundary markers: The skill uses structural headers like ## Code to Analyze to delimit code, but lacks robust techniques like randomized delimiters to prevent malicious instructions within analyzed files from overriding sub-agent behavior.
  • Capability inventory: The orchestrator can execute shell commands and leverage sub-agent tools (like file-writing or network access) if provided by the environment.
  • Sanitization: No explicit sanitization or escaping is performed on the content of the files before they are passed to the sub-agents.
Audit Metadata
Risk Level
SAFE
Analyzed
May 12, 2026, 04:30 PM
Security Audit — agent-trust-hub — multi-ai-collab