What the agent does
This is for a software development agency that builds apps for several clients at once and tracks each client's work as a project in Linear. Every Friday each account manager writes the client an update: what shipped, what is in progress, and what the team is waiting on from the client. The facts are all in Linear, but in the developers' words ("Fix race in token refresh", "Spike: offline sync"), and turning forty issue titles into five sentences a client's operations manager understands takes an hour per client. The agent writes the first draft. The run names one connection, linear, signed in once by a lead on the agency's Linear workspace with POST /me/connections {"type": "app", "name": "linear"}. Rush hands that sign-in to the agent as Linear's own MCP server.
For each client project named in the input, the agent reads the issues completed in the last seven days, which in Linear's terms are issues whose workflow state type is completed and whose completedAt falls in the week, and the issues whose state type is started. It reads each issue's title, description, labels, assignee and the comments posted this week, because the comment thread is usually where the real status lives ("merged, waiting on App Store review"). It skips issues carrying the label the agency uses for internal work, internal in this request, so refactors and tooling never reach a client.
Issues in the state the agency uses for client blockers, or labeled needs-client, become the "waiting on you" section. For each one the agent writes the question the client has to answer in one sentence and the date of the last comment from the agency's side, so the account manager can see how long the client has had it. It also reads the project's target date and says plainly whether the week's progress leaves it on track, slipping or unknown, without inventing an estimate.
It writes one Markdown file per client, updates/<client>.md, in the agency's tone from the input: a two-line summary, Shipped, In progress, Waiting on you, and Next week, each item in the client's language with the Linear issue identifier kept in brackets for the account manager. It changes nothing in Linear. Linear's server can create and update issues and comments; the prompt forbids every write, and the agent sends nothing to any client.
Start it with one API call
POST https://api.phoenix.so/o/{org}/sessions with this body:
{
"agent": "claude",
"input": "Use the Linear connection. For each of these Linear projects: Harbor Mobile App, Fieldnote Web Portal, Pantry Ops Dashboard, read the issues whose workflow state type became completed in the last 7 days (completedAt) and the issues whose state type is started. Skip any issue labeled internal. Read each issue's title, description, labels, assignee and this week's comments. Issues labeled needs-client or in the Blocked state go under 'Waiting on you': write the client's question in one sentence and the date of our last comment on it. Read the project's target date and say on track, slipping or unknown, with the reason. Write updates/<project name>.md for each project: a two-line summary, then Shipped, In progress, Waiting on you, Next week. Write for a non-technical client in a friendly, direct tone, no jargon, and keep the issue identifier in brackets after each item. Do not create, update or comment on anything in Linear, and do not contact any client.",
"connections": [
"linear"
]
}- Get an API key. In the console, open Settings, then API keys, and create a key for your org. Send it in the Authorization header as a Bearer token.
- Send the request. POST the body in request.json to https://api.phoenix.so/o/{org}/sessions, with your repository as the resources URL. The run starts in its own cloud sandbox.
- Read what came back. Follow the run on its live stream or wait for the run.completed webhook, then read the result and its trajectory.
What comes back
The session returns one Markdown file per client as artifacts, each with a download url, and a short list in result of which projects had nothing shipped this week. Each account manager opens their client's file, edits it, and sends it from their own email or the client's shared channel. The bracketed issue identifiers are for the account manager; they delete them or keep them, depending on whether the client reads Linear.
What the account manager checks before sending: that nothing labeled internal slipped through under another label, that a "shipped" item really reached the client (completed in Linear can mean merged but not yet released), and the on-track sentence, which the agent bases only on the target date and the issues it read. They also check the "waiting on you" questions, since a client who gets asked the same thing twice notices. The transcript shows each Linear read, so a missing item can be traced to its state or label.
Run it on every Friday afternoon
In the console, a Project automation with a schedule trigger, cron 0 14 * * 5 in the agency's timezone, runs a saved agent whose instructions hold the agency's tone, the internal label and the client-blocker state, and whose connections name linear. The automation's prompt lists the active client projects, so adding a client is one line.
An agency that keeps one Linear team per client can name teams instead of projects in the same prompt. Each Friday is its own session, so the drafts from earlier weeks stay readable and a client's history of updates can be read in order.
The use case this job follows, with its automation and trigger: Produce a report or file, no repo.
Why run it on Rush?
Linear's own project updates are written by a person, and an export of completed issues reads like a changelog, not a letter to a client. The agent reads the comments as well as the titles, which is where "done" and "done but waiting on review" differ, and it writes in the agency's voice for someone who has never seen the board. Linear's sign-in stays in Rush's connection, never as an API key in a script.
Each run has its own computer and a record of every read and its cost. When a client asks why an item was reported as shipped, the account manager can open that Friday's run and see the issue and comment it came from.
Frequently asked questions
Will the agent send the update to the client?
No. It writes drafts and stops. The account manager reads, edits and sends each one, because an update is a promise to a client and a person makes it.
Can it post the draft back to Linear?
Linear's server can write comments and update issues, so the prompt could allow it, for example a comment on the project's tracking issue. This request forbids every write so a draft never lands where a client might see it before a person has read it.
We track some clients in teams, not projects. Does that work?
Yes. Name the teams in the prompt instead of the projects; the agent reads the same issues, states and labels either way.
How does it avoid leaking internal work?
It skips every issue with the internal label the prompt names. An internal issue that was never labeled is the gap, which is why the account manager reads the draft first.