Custom specialist agents with GSD — what's the recommended integration pattern?

[ismaHenzel]

I've been using GSD for a while and it's been great for long-running, structured development. But I'm hitting a gap when it comes to custom specialist agents.

My setup has custom agents like:

• 1. KB-driven code agents — e.g. a python-developer agent that reads from .claude/kb/python/ before writing any code, enforcing project-specific patterns, dataclass
     conventions, type hint rules, etc.
• 2. Tool-heavy agents — agents that carry specific MCP servers or custom tools that the generic gsd-executor doesn't have access to (e.g. a database agent with a Postgres MCP,
     or a cloud agent with provider-specific tools).

The gsd-executor does the heavy lifting of plan execution, commits, summaries and state management — and I don't want to lose any of that. But for certain tasks (e.g. all .py
files in a task), I'd like the executor to delegate the actual implementation to the right specialist, then resume to handle the commit + SUMMARY.md as usual.

My current thinking is:

• Add Agent to the executor's tools list
• Add a <specialist_routing> section that maps task file patterns to agent types
• The specialist writes the files, the executor picks up and commits

A few questions:

1. Is this the intended extensibility path to inject custom agents?
2. If agent delegation is valid, how should the specialist signal completion to the executor — just by having written the files, or is there a structured handoff format?