multi-agent-coordination
Pass
Audited by Gen Agent Trust Hub on Feb 27, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [COMMAND_EXECUTION]: The skill uses shell commands to invoke a CLI tool named 'claudish' for proxying requests to external AI models. Examples include: 'RESULT=$(claudish --model x-ai/grok-code-fast-1 --stdin <<< "$PROMPT")'.
- [PROMPT_INJECTION]: The skill demonstrates a vulnerability surface for indirect prompt injection (Category 8). It implements patterns where an orchestrator reads content from project files (e.g., 'ai-docs/architecture-instructions.md', 'src/auth.ts') and interpolates this data into prompts for sub-agents without explicit sanitization or boundary markers. Evidence includes: 1. Ingestion points: Files read from 'ai-docs/', 'src/', and 'tests/' directories. 2. Boundary markers: Absent in the provided prompt templates. 3. Capability inventory: File system read/write access and shell command execution ('claudish', 'bash'). 4. Sanitization: No sanitization or validation of the ingested file content is described.
- [EXTERNAL_DOWNLOADS]: The skill facilitates communication with external AI providers, specifically 'x-ai' (Grok) and 'google' (Gemini), via a non-standard CLI tool. While these are well-known services, the use of an external CLI tool to transmit data from the workspace is a significant network operation.
Audit Metadata