Examples

Ideas you can try right now

Already running Frontier? Try out a few examples to figure out new ways to automate your daily routine, and earn back your life.

GitHub Issue Pipeline

GitHub

A scheduled agent polls your open issues and reproduces the ones it can. Each reproducible issue gets its own workspace and branch, where a Builder writes the fix and a Reviewer checks it before opening a real pull request — so two fixes in flight never collide.

Before you try it

  • A GitHub repo you can push to
  • The `gh` CLI, signed in (`gh auth login`)

The team

  • Issue ScoutPolls your real GitHub issues with `gh`, reproduces what it can in this workspace's own checkout, and writes up what it found.
  • Scout Build ReviewThe whole pattern, saved once — your next repo gets the same team in minutes.
  • BuilderFixes one issue, in its own workspace, with a test. Comments the branch back onto the real GitHub issue.
  • ReviewerPushes back until it's satisfied, then opens the real pull request and links it from the issue. Nothing merges without you.
  • BugfixesWhere the Scout lives. Every issue it reproduces gets its own child workspace and branch, so two fixes in flight never collide.

How it runs

  1. The Scout finds a new issue overnight and reproduces it against your actual code.
  2. It spins up a fresh workspace just for that fix, with its own branch, and hands it to a Builder.
  3. The Reviewer reads the diff, asks for a regression test, and opens the real PR once satisfied.
  4. You wake up to a finished pull request and a comment trail on the original issue — read it, merge it, done.

Built with

  • Scheduled agent
  • Strategy
  • 2 Watcher agents
  • Workspace

One workspace per issue means concurrent fixes never fight over the same branch — a Builder is twice as useful with a Reviewer attached, and disagreement between agents is quality control you don't have to perform yourself.

Jira Bug Pipeline

JiraGitHub

The same pipeline as the GitHub Issue Pipeline, but the tracker is Jira Cloud. A scheduled agent finds new bugs through Atlassian's own MCP server, reproduces them against your repo, and hands reproducible ones to a Builder and Reviewer that link the finished pull request back onto the ticket.

Before you try it

  • A Jira Cloud site (one-time `claude mcp add` + browser login to the Atlassian MCP server)
  • The GitHub repo the ticket's code actually lives in, with `gh` signed in

The team

  • Jira ScoutCalls the Atlassian MCP server's Jira tools to find new bugs, reproduces what it can against the real repo, and comments progress back onto the real ticket.
  • BuilderFixes the bug with a test, in its own branch, and comments the branch back onto the Jira ticket.
  • ReviewerReviews the diff, opens the real pull request, and links it from the Jira ticket. Nothing merges without you.
  • Jira BugfixesWhere the Scout lives. Every bug it reproduces gets its own child workspace and branch.

How it runs

  1. A bug is filed in Jira. The Scout finds it on its next pass and reproduces it against real code.
  2. It spins up a dedicated workspace for the fix and comments "in progress" back onto the ticket.
  3. The Builder and Reviewer work the fix the same way Issue Triage does, entirely in GitHub.
  4. The finished pull request links back to the Jira ticket, so both tools stay honest about where things stand — read it, merge it, done.

Built with

  • Scheduled agent
  • 2 Watcher agents
  • Workspace

The tracker your team files bugs in doesn't have to be the tool your agents work in — an official MCP server carries the connection, and the same Builder/Reviewer pattern moves the fix regardless of where the report came from.

Release Notes Drafter

GitHub

A scheduled agent lists every pull request merged since the last tag, groups them into Added, Fixed, and Changed, and drafts the changelog for you — into a file in the repo and into the thread, ready to cut.

Before you try it

  • A GitHub repo with tagged releases
  • The `gh` CLI, signed in

The team

  • Release DrafterFinds every pull request merged since the last tag and turns real titles and diffs into a real draft — nothing invented.
  • Release NotesAttach the real repo here. The draft and its history live in this workspace, not in your head the night before a release.

How it runs

  1. The Drafter wakes on its schedule and checks what's merged since the last tag.
  2. It groups the real pull requests into Added, Fixed, and Changed, and writes the draft into a file in the repo.
  3. It posts the same draft into the workspace thread.
  4. You read it, touch up anything that needs a human voice, and cut the release — that's the whole job left.

Built with

  • Scheduled agent
  • Workspace

A changelog written from the actual commit history is more accurate than one written from memory, and a draft that's already 90% done takes less of your evening than a page you fill from nothing.

Support Inbox Triage

EmailZendesk

A scheduled agent checks a dedicated support mailbox over IMAP and drops each new message into its own thread here. A watcher agent reads it from there — filing a real Zendesk ticket for a bug or feature request, or drafting an answer ready to send for a question.

Before you try it

  • A support mailbox and an app password (Gmail, Outlook, or any IMAP provider)
  • A Zendesk instance and an API token (Admin Center → Apps and integrations → Zendesk API)

The team

  • Email AccessA read-only IMAP login for the mailbox — the Scout can read it, never send, delete, or move anything.
  • Zendesk AccessA Zendesk API token, so Support Triage can file real tickets.
  • Inbox ScoutChecks the mailbox for new messages and starts a thread for each one — the same shape as pasting it in yourself.
  • Support TriageReads each message and either files a real Zendesk ticket, or drafts an answer ready to send — never both, and never sent without you.
  • Support InboxWhere every message lands, one thread each — from the Scout, or pasted in by hand.

How it runs

  1. A message lands in the support mailbox.
  2. The Inbox Scout finds it on its next pass and starts a thread for it here — no paste required.
  3. Support Triage decides: bug, feature request, or answerable question.
  4. A bug or request becomes a real Zendesk ticket with the link back to you; a question gets a drafted answer instead — read it, and send.

Built with

  • 2 Domains
  • Scheduled agent
  • Watcher agent
  • Workspace

Triage is a decision, not a task — once the mailbox feeds itself in, the agent makes the call and files the real record or drafts the reply, so the only thing left for you is hitting send.

Dispute Failed Payments

Stripe

A scheduled agent watches Stripe, through a read-only key, for new failed payments and disputes. It drafts a dunning message for one and a summary for the other — every draft lands in the workspace for you to read, and nothing goes out or gets refunded without you.

Before you try it

  • A Stripe account and a restricted, read-only API key

The team

  • Stripe AccessA read-only key, scoped to just two read tools, registered with Stripe's own MCP server — nothing here can move money.
  • Billing WatchdogCalls Stripe's MCP tools for new failed payments and disputes, and drafts what to say about each.
  • BillingWhere the drafts land — dunning messages and dispute summaries, ready for you to read.

How it runs

  1. A charge fails or a dispute opens in Stripe.
  2. The Watchdog finds it on its next pass and drafts a direct message for the customer, or a summary if it's a dispute.
  3. You read the draft in the workspace, touch up the tone if you want to, and send — that's the whole job.

Built with

  • Domain
  • Scheduled agent
  • Workspace

A read-only key is enough to turn "I found out too late" into "I saw it the same morning" — the agent watches and drafts, and every decision that touches money stays yours.

Production Error Pipeline

HoneycombGitHub

A scheduled agent watches your OpenTelemetry traces in Honeycomb for exceptions that are new or spiking against their own baseline, reproduces what it can, and hands anything reproducible to the same Builder/Reviewer pipeline as the GitHub Issue Pipeline — one workspace and branch per error.

Before you try it

  • A Honeycomb account with OpenTelemetry traces already landing in a dataset, and a read API key
  • The GitHub repo the service's code lives in, with `gh` signed in

The team

  • Honeycomb AccessA read API key and dataset name, so the Scout can query your real traces.
  • Error ScoutQueries production traces for exceptions new or spiking against their baseline, reproduces what it can, and hands each real one to its own fix workspace.
  • BuilderFixes one error signature with a test, in its own branch, keeping the original trace link attached to the work.
  • ReviewerReviews the diff and opens the real pull request. Nothing merges without you.
  • Production ErrorsWhere the Scout lives. Every error it reproduces gets its own child workspace and branch.

How it runs

  1. A new exception type starts showing up in production traces, or an existing one spikes.
  2. The Error Scout notices it on its next pass, pulls a representative trace, and tries to reproduce it against the real repo.
  3. A reproducible error gets its own workspace and branch; the Builder fixes it, the Reviewer checks it and opens the pull request.
  4. You read a pull request whose description links straight back to the trace that caused it to exist — merge it, done.

Built with

  • Domain
  • Scheduled agent
  • 2 Watcher agents
  • Workspace

A stack trace and a GitHub issue describe the same bug in different vocabularies — the same reproduce/fix/review pipeline works on either, once something translates between them.

Web Vitals Monitor

HoneycombGitHub

A scheduled agent compares each route's Core Web Vitals against its own rolling baseline and, for anything that regressed, lists what actually merged into that route's code in the same window. It diagnoses; it never touches code.

Before you try it

  • A Honeycomb account with browser OpenTelemetry traces (Core Web Vitals) already landing in a dataset, and a read API key
  • The GitHub repo the frontend's code lives in, with `gh` signed in

The team

  • Honeycomb AccessA read API key and dataset name, so the Monitor can query your real browser traces.
  • UX MonitorCompares each route's Core Web Vitals against its own rolling baseline and names the pull requests most likely responsible for a regression.
  • UX MonitoringWhere the report lands, one day at a time — the numbers and the suspect pull requests, side by side.

How it runs

  1. A route's p75 largest-contentful-paint (or another vital) creeps past its own baseline.
  2. The Monitor notices on its daily pass and lists what merged into that route's code in the same window.
  3. It posts a ranked report in the workspace and opens a real GitHub issue tagged "performance" with the same findings.
  4. You read a short list of suspects instead of starting a performance investigation from zero.

Built with

  • Domain
  • Scheduled agent
  • Workspace

A regression an agent can name a suspect for is a five-minute conversation; one that arrives as "the app feels slow" is an afternoon — the difference is whether anything was watching the metric before someone complained.

Roadmap Dispatcher

A checklist in a workspace prompt is the roadmap. A scheduled Dispatcher hands the topmost untouched item to a Lead Engineer to scope and build, and a Verifier checks it off only after actually running the finished result — the agent that builds something never grades its own work.

The team

  • CuratorPeriodically rereads the mission, the finished work, and the feedback — and proposes what belongs on the list next.
  • DispatcherWakes on its interval, finds the topmost untouched item, and hands it to the lead engineer workspace. Caps how much is in flight at once, and nudges stalled threads instead of piling on new ones.
  • VerifierRuns the finished work, and only then checks the item off the roadmap and archives it.
  • Lead EngineerScopes each item into a plan, then breaks it into child workspaces — one per deliverable, each with its own thread and worktree.
  • MissionThe plain-language mission at the top, then the roadmap as a checklist. Items carry status tags as they move — untouched, in flight with a link to their thread, done.

How it runs

  1. You add a line to the checklist and go to your meetings.
  2. The Dispatcher hands it off and tags it with its thread, so anyone can see where it stands.
  3. The lead engineer scopes it; child workspaces build it; the Verifier proves it works and checks it off.
  4. The Curator notices what the finished work makes possible and drafts the next three items for your review.

Built with

  • 2 Scheduled agents
  • Watcher agent
  • 2 Workspaces

One agent decides what's next, others do the work, and a separate one confirms it's done — never the same agent grading its own homework.

Persona Product Reviews

Each user persona — the hurried one, the skeptic — is its own workspace, written as a character study. A scheduled agent hands every persona the same review assignment on a cadence you set, and a Synthesizer merges their findings into one ranked list.

The team

  • Review CycleOn a cadence you choose, fans a review pass out to every persona workspace: walk the product's main journeys as this person, and say what grated.
  • Product LeadWhere the findings converge — one thread per cycle, duplicates merged, the sharpest complaints on top.
  • VoicesThe parent room: who your users are, and one child workspace per persona beneath it, each prompt a character study — their goals, their patience, their deal-breakers.

How it runs

  1. The Review Cycle fires on Friday and hands each persona the same assignment.
  2. The skeptic gets lost on step two; the power user wants a shortcut; the hurried one gives up at the signup form.
  3. Their findings land with the product lead, synthesized into a ranked list.
  4. Monday's roadmap conversation starts from evidence instead of instinct.

Built with

  • Scheduled agent
  • 2 Workspaces

Perspective is a prompt — a handful of well-written personas will find what you've gone blind to, on schedule.

Quote Follow-Up

Email

A scheduled agent checks your sales inbox over IMAP and starts a thread for each new quote request. A watcher agent drafts the estimate from there, and if the customer goes quiet, nudges again a few days later instead of letting the request go cold.

Before you try it

  • A sales inbox and an app password (Gmail, Outlook, or any IMAP provider)

The team

  • Email AccessA read-only IMAP login for the sales mailbox — the Scout can read it, never send, delete, or move anything.
  • PricingWhat you charge for what — line items, ranges, and the caveats worth stating up front, so drafts stay consistent with each other.
  • Follow-UpChecks threads with a sent quote and no reply after a few days, and drafts a short nudge instead of leaving it to go cold.
  • Inbox ScoutChecks the mailbox for new messages and starts a thread for each request that looks like a job.
  • EstimatorDrafts the quote from the real request and the Pricing domain — ready to review and send, never sent for you.
  • QuotesWhere every request lands, one thread each — from the Scout, or pasted in by hand.

How it runs

  1. A "how much would it cost to..." email lands in the sales inbox.
  2. The Inbox Scout finds it on its next pass and starts a thread here — no paste required.
  3. The Estimator drafts a quote from the real request and your stated pricing; you read it, adjust anything that needs a human touch, and send.
  4. No reply after a few days? Follow-Up drafts a short nudge for that thread instead of the request quietly dying in your sent folder.

Built with

  • 2 Domains
  • 2 Scheduled agents
  • Watcher agent
  • Workspace

Most lost jobs aren't lost on price — they're lost to a slow first reply or a follow-up nobody got around to. An agent that drafts both keeps the response time short without you watching the inbox all day.

Overdue Invoice Nudge

Stripe

A scheduled agent watches Stripe, through a read-only key, for invoices that have gone past due. It drafts a short, direct payment reminder for each one — every draft lands in the workspace for you to read, and nothing goes out without you.

Before you try it

  • A Stripe account issuing invoices, and a restricted, read-only API key

The team

  • Stripe AccessA read-only key, scoped to just the invoice-read tools, registered with Stripe's own MCP server — nothing here can move money.
  • Invoice WatchdogCalls Stripe's MCP tools for invoices past due and drafts a reminder for each, tuned to how overdue it is.
  • InvoicingWhere the drafts land — one thread per overdue invoice, with the amount and the client attached.

How it runs

  1. An invoice passes its due date unpaid in Stripe.
  2. The Watchdog finds it on its next pass and drafts a reminder — polite the first week, more direct if it's been a month.
  3. You read the draft in the workspace, touch up the tone if you want to, and send — that's the whole job.

Built with

  • Domain
  • Scheduled agent
  • Workspace

Chasing payment feels awkward enough that it gets put off, which is exactly how invoices go from late to unpaid — a drafted reminder removes the friction of starting, while the decision to send stays yours.

Reschedule Assistant

Email

A scheduled agent checks your booking inbox over IMAP and starts a thread for each new message. When a customer emails to cancel or asks to move their time, a watcher agent drafts a reply pointing them back to your booking link — instead of that email sitting unanswered until someone notices.

Before you try it

  • A booking inbox and an app password (Gmail, Outlook, or any IMAP provider)
  • Your booking link or phone number, so drafts can point customers to it

The team

  • Email AccessA read-only IMAP login for the booking mailbox — the Scout can read it, never send, delete, or move anything.
  • Inbox ScoutChecks the mailbox for new messages and starts a thread for each one.
  • Reschedule DeskReads each thread; a cancellation or reschedule request gets a drafted reply pointing back to your booking link, ready to send.
  • BookingsWhere every booking-related message lands, one thread each.

How it runs

  1. A customer emails to say they can't make their appointment.
  2. The Inbox Scout finds it on its next pass and starts a thread here.
  3. The Reschedule Desk recognizes the request and drafts a short reply with your booking link, so they can pick a new time themselves.
  4. You read the draft, send it, and the slot gets rebooked without a phone call.

Built with

  • Domain
  • Scheduled agent
  • Watcher agent
  • Workspace

A missed appointment costs less when the seat gets rebooked fast — the agent can't see your calendar, but it can make sure a reschedule request never just sits in the inbox waiting for someone to notice it.

Lead Intake Triage

Email

A scheduled agent checks the inbox your contact form sends to and starts a thread for each new lead. A watcher agent sorts it — a real job in your service area gets a fast personal reply drafted, and anything out of area or clearly spam gets set aside instead of clogging the same inbox as real customers.

Before you try it

  • The inbox your contact form or "request a quote" button sends to, and an app password

The team

  • Email AccessA read-only IMAP login for the leads mailbox — the Scout can read it, never send, delete, or move anything.
  • Service AreaWhat you actually take on — service area, job types, minimums — so triage can tell a real lead from a bad fit.
  • Inbox ScoutChecks the mailbox for new messages and starts a thread for each inquiry.
  • Lead TriageReads each inquiry against the Service Area domain and drafts a personal reply for a real lead, or a short decline for a bad fit — never sent without you.
  • LeadsWhere every inquiry lands, one thread each, sorted as triage decides.

How it runs

  1. Someone fills out the contact form, or emails in directly.
  2. The Inbox Scout finds it on its next pass and starts a thread here.
  3. Lead Triage checks it against what you actually take on and drafts a fast, specific reply — or a polite decline if it's not a fit.
  4. You read the draft and send it, often before the lead has finished checking other options.

Built with

  • 2 Domains
  • Scheduled agent
  • Watcher agent
  • Workspace

Response speed wins jobs as often as price does — an agent that reads and drafts the moment a lead lands means the fastest reply doesn't depend on who happened to check email first.

The parts are the same in every story. The arrangement is yours.