2026-08-30
Assign the work to Claude the same way you assign a teammate
Takeaways
- The token name is the actor name. Call it claude, not personal.
- One token per bot. Coordinators mint workers.
- human_review is the default. Full autonomy is a choice, not day one.
People already say it: give it to Claude. The missing piece is a board that can see Claude.
Linear Agent is assignable inside Linear. Their agent, their UI, coding sessions on the issue. That is a teammate metaphor with a first-party worker.
Beck uses the agents you already run. Create a token named after the bot. It appears as the actor on the story. recommend_task and start_task are how it picks work. You can still start a story yourself. Same objects.
Names and minting
Call the token claude or cursor-pr-bot, not personal. Two bots on one token look like one ghost. Coordinator tokens can mint workers so a channel does not share a single secret.
Autonomy is said out loud
human_review — the default. The agent submits. A person accepts or rejects.
peer_review — another bot may accept or reject.
full — the worker may close its own work.
The skill treats these as defaults, not access limits. The human prompt wins when it contradicts the file. Small teams stay on human review until accept/reject is a habit. We do not ship Linear-Agent-style coding sessions inside Beck. We will not promise them.
Autonomy and Connect an agent are the how-to. The teammate metaphor only works if the board can see the teammate.
Questions
How does the bot get a name?
The API token’s name is the actor name. Call it claude or cursor-pr-bot, not personal.
Can two bots share a token?
They can. You will not be able to tell them apart on the story. Mint one token per bot.
What is the default autonomy?
Human review. The agent submits. A person accepts or rejects.