other-agents

When Linear Agents was one umbrella, it was holding too many ideas. The coder-agent earned its own project. What’s left becomes this one — everything that isn’t write code for me.

First candidate: triage-agent. A non-coding agent that watches the triage team, reads incoming issues, and classifies them against a cached taxonomy (project, priority, labels, likely assignee). Elicitation threshold so it asks me when confidence is low rather than guessing. Deployed at triage-agent.gabi.tv, same pattern as the coder-agent — Coolify application, Linear OAuth install, webhook on Issue.create.

After that, others arrive when the itch appears. A weekly-review agent. A research-assistant agent that takes an open question, runs it, and comes back with a document. A “factorise the shared scaffold” ticket sits here too — when three agents exist in production, the copy-paste pattern probably earns a shared library.

The rule stays the same as with the coder-agent: if the agent can do it end-to-end without me holding its hand, it stays. Otherwise it gets killed.