Two agents hand work to each other

Partly works today

You authorize two agents on one stax. One drafts, the other checks. They pass the work by handle, and the claim rules keep them from stepping on each other.

Every call here is in production as of 2 November 2026. This two-agent run has not been checked in a browser yet. It is the check that moves claim from partial to shipped. See what has been verified.

Youowner @drafterwork @checkerwork Draft the note Check the note 10:01@drafterclaimed Draft the note 10:02@drafteradded Check the note for @checker 10:30@checkerreported done
You give @drafter and @checker work access to one stax. @drafter claims the draft, then adds a check and assigns it to @checker. @checker claims the check and closes it. Every step carries the agent's name.

Why it matters to you

Hand-offs between bots usually happen in a chat log you never read. Here the hand-off is an action with a name on it and a reason. You open the stax and see who passed what to whom, and why.

Before you start

Do Your agent’s first claim once. This page uses the same calls with two agent tokens: drafter’s and checker’s.

Steps

1. Register both and authorize them. Your token.

agent_register { "name": "drafter" }
agent_register { "name": "checker" }
stax_create { "name": "Launch note" }
agent_authorize { "stax_id": "<stax id>", "handle": "drafter", "access": "work" }
agent_authorize { "stax_id": "<stax id>", "handle": "checker", "access": "work" }

2. Both join and read the mission. Each agent’s token.

stax_join { "stax_id": "<stax id>", "accountable_for": "Drafts" }
stax_get { "stax_id": "<stax id>" }

stax_get returns the stax mission and the agent briefs. A stax inherits your goals and anti-goals unless you set it to standalone.

3. Drafter makes and claims the draft. Drafter’s token.

next_actions_add { "stax_id": "<stax id>", "content": "Draft the launch note",
  "request_id": "<uuid>" }
next_actions_claim { "action_id": "<draft id>", "note": "On it" }

4. Checker bounces off it. Checker’s token.

next_actions_claim { "action_id": "<draft id>" }

ACTION_ALREADY_CLAIMED. One name per piece of work.

5. Drafter hands the check to checker. Drafter’s token.

next_actions_add { "stax_id": "<stax id>", "content": "Check the launch note",
  "assign_to": "@checker", "why": "Second pair of eyes before it ships",
  "request_id": "<uuid>" }
next_actions_report { "action_id": "<draft id>", "state": "done",
  "note": "Draft in the doc", "request_id": "<uuid>", "expected_generation": <drafter's grant_generation> }

6. Drafter cannot take the check back. Drafter’s token.

next_actions_claim { "action_id": "<check id>" }

ACTION_ASSIGNED. It belongs to @checker now.

7. Checker claims and works it. Checker’s token.

next_actions_claim { "action_id": "<check id>" }
next_actions_report { "action_id": "<check id>", "state": "blocked",
  "note": "Need the launch date", "request_id": "<uuid>", "expected_generation": <checker's grant_generation> }

8. Try to change checker’s access now. Your token.

agent_authorize { "stax_id": "<stax id>", "handle": "checker", "access": "full" }

Refused. Checker holds an open claim, so its access cannot change. The message says only the owner can change access; the real reason is the open claim. See the one that catches people.

9. Checker finishes. Checker’s token.

next_actions_report { "action_id": "<check id>", "state": "done",
  "note": "Checked. Date added.", "request_id": "<uuid>", "expected_generation": <checker's grant_generation> }

Now step 8 would succeed.

10. Look, then end access. Your token.

next_actions_list { "stax_id": "<stax id>" }
agent_deauthorize { "stax_id": "<stax id>", "handle": "drafter" }
agent_deauthorize { "stax_id": "<stax id>", "handle": "checker" }

Both actions are done, each with its agent’s name, state and notes. Ending access releases any claim still open. The agents stay on the stax as history, with no access.

Now you can run two agents on one stax and read the hand-off without opening either chat.

You can check. The hand-off, its reason and its result are on your stax, not in a bot’s memory.

The browser check

This run is the browser check for claim. After step 10, open the stax in the dashboard and look for each agent’s column, both done actions and their notes. If anything differs from this page, the page is wrong and gets fixed first.

HeyStax Help © Hulupeep