multi-agent-coordination

This document is a structured orchestration guide describing patterns for coordinating multiple AI agents, delegating via files, and invoking external model CLIs. The document itself is not executable code and contains no direct malicious payloads, hardcoded secrets, or obfuscated code. The primary security concerns are operational: implementers who follow the patterns could accidentally forward sensitive workspace contents or credentials to external models/CLIs (credential forwarding/data exfiltration), or allow prompt-injection through untrusted files passed between agents. Risk mitigations are absent in the text (no redaction, allow-listing, or explicit warnings about excluding secrets). Recommended mitigations for anyone implementing these patterns: enforce allow-lists for files sent to external models, sanitize/redact secrets from files, require explicit user confirmation before sending workspace contents externally, log all external calls, and minimize the set of files created/read during preparation steps. Overall, the material is conceptually useful for orchestration but operationally requires careful controls when implemented to avoid data leakage.