Work from your own tools · Lesson 5 of 5
How do you ask @yagni for help?
Published June 9, 2026 · Updated September 25, 2026
One rail, one voice, scoped to what is in the center. On Work it reads across your Teams and starts tickets; on a case file it knows the plan, the review, and the decisions. It answers from the decision ledger, your Teams' record, and the rows of your connected tools, and speaks for the Agents, never instead of them.
Where do you talk to @yagni?
In the right rail, on every surface, scoped to what is in the center. Open Work and it reads across your Teams. Open a case file and it is already on that ticket, its plan, its review, its decisions. Open a Team page or a line and it knows that Team’s scope and what each line is set to. Same voice, same memory, everywhere. The only difference is how much context comes with the question.
What can it start?
Tickets. Its one creating verb is “start a ticket for”: describe a change that is not filed yet, and it starts on Work, in Plan, credited to you. It is not filed in your tracker. Paste a Linear or Jira link or key and the ticket lands under Proposed for you to accept. Paste a list and each item lands there quietly. It never writes a ticket form for you, because there is none.
What grounds the answers?
Three things, and every answer says which. The decision ledger, so “why queues instead of cron” comes back with the decision, who made it, and what it ruled out. Your Teams’ own record, so “how did that plan get approved” comes back with the decisions and who answered them. And the live rows of your connected tools, so “is the webhook fix merged” is answered from GitHub rather than from memory.
When there is no position on record, it says so and names the assumption it is proceeding on rather than dressing a guess as a fact.
Can you ask why an Agent did something?
Yes, and it is the question worth asking most. Ask why Reeve flagged a plan, why Proctor held a review open, why Bailey proposed this ticket ahead of that one, and you get the reasoning with the evidence attached: the rule that fired, the ledger entry it read, the thread where the scope changed.
That is also where a correction starts. If the answer is wrong, saying so on the case file is what teaches the Team, and the Playbook rule that comes out of it is the Team’s, not one person’s. More on that in when an Agent gets it wrong.
What does it never do?
Decide for you. The footer of the case file is the only place a decision is answered, and it is answered by the person it is addressed to. @yagni explains the decision, shows the evidence, and drafts the note you might send back, but the click is yours. It also never signs its name to a Team’s pull request. When you read “Wright opened this on Checkout”, that is the whole point.
Where do you go next?
The same grounding rides in the terminal: YAGNI Code reads the ledger before it plans, and a harness connected over the YAGNI MCP server can ask the same questions. For who is behind each answer, read the Agents.
Common questions
Is the rail on a case file a different agent from the rail on Work?
No. One voice with one memory. What changes is the context in front of it: on Work it reads across your Teams, on a ticket it knows that ticket, on a Team page it knows that Team and its lines. There is no separate chat page or thread list; the rail is the whole conversation.
If there is one voice, who does the work?
The Agents on your Teams. @yagni speaks for them, explains their work, and takes requests, but the work is credited to the Agent that did it on the Team it served. One voice in the rail, many names on the record. @yagni's own name appears on work only where no Team exists yet.
Can @yagni approve a plan or merge for me?
No. It never replaces a form or a footer. The decisions a ticket needs resolve on the case file, in the footer, by the person they are addressed to. What the rail does is explain the decision in front of you, with the evidence, so you answer it faster and better.
Will it just agree with me?
No. You get its read and how confident it is. When the evidence cuts against you it shows you the evidence, and when the call is genuinely close it says so and hands it back. When there is no position on record it names the assumption it is proceeding on rather than dressing a guess as a fact.