Custom agents usage with spec-kit?

[benko325]

Hi, I am using Claude Code and I have already created some custom agents before (fe, fe code reviewer, be, specified be, be code reviewer, test writer, ...) and I am wondering if there is a batter way to use them than just do something like "/implement task t001 with be-agent" or there is more "clever" way.

[willcooley]

@benko325 not sure if this helps, but he is a more structured process I came up with you can try out?

1. Use DollhouseMCP, and create your own custom agents for your projects; like frontend-engineer, backend-engineer, database-engineer, and etc. (create repository code specialist agents, and not work type generalist)
2. Export out and convert those files into agents for Anthropic if you are using Claude, or custom import them into your ide/ cli as a custom defined agent to use
3. Update your Spec-kit template files to coordinate planning, spec, and tasks based on agent and agent's domain. If needed, you should be able to create custom slash commands as well for further steps
4. Have the agents contain criteria to only work on a specific tasks requirement that has a designation; like provide guardrails.
5. Spawn a separate terminal for each agent to work on their own designated tasks groups from the document, and instruct them not to overlap; this prevent context bleeding while them /speckit.implement work. If you did create custom slash commands, you can do something like /speckit.implement-db, /speckit.implement-fe, /speckit.implement-devops, /speckit.implement-be

This process allows you to execute tree branching work execution to expedite development process vs a sequential execution of tasks by one agent; so it's exponentially faster, but does require more CPU usage and tokens.

Time Savings: 50-60% faster development with parallel agent execution