[FEATURE] One Claude Instance Directly Injects Prompts into Second Instance

https://github.com/parruda/claude-swarm/issues/115#issuecomment-3210868717

Would be great if I could run a second instance, called a judge (or guide), that would watch what the first instance is doing, and if it notices something not working as instructed then inject a prompt directly into the first instance, instead of having to use something like MCP.

That same way a human would do so.

The payload names no repository files, tests, or entry points. Start by reviewing the linked claude-swarm issue comment and clarifying the judge/guide interaction before locating the relevant TypeScript agent-orchestration entry point. Done should mean a second instance can observe the first and inject a prompt when it detects a deviation.

Written by the indexing model from the issue text.