Mr Fernby, a goat in a butler's waistcoat, collar and monocle

Mr Fernby

Your project's knowledge, now accessible to you and your agents.

You know when documentation goes out of date, right? The moment it's written.

Agents produce work faster than a person can review it, so the person becomes the bottleneck, and eventually reviews just get skipped. Mr Fernby is the fabric that makes the person's attention go further. Docs and tickets in one place. The right context for every task. Corrections that stick. Parallel sessions that don't collide.

You've written this prompt.

Probably this week.

Normal agent prompts

C:\dev\ShoppingBeast> agent

agent 3.1.0

cwd: C:\dev\ShoppingBeast

This is the checkout service. It's an API with a web front end, and an order goes through three services: cart, pricing and ledger. Pricing owns discounts, NOT cart. Read every markdown file in the repo before you do anything.

Some of the markdown is out of date. Check when a file was last edited before you rely on it, and if it's more than a few weeks old, check it against the code. docs/refunds.md is wrong, ignore it.

Use the new logger, not the old static one. I have told you this before. Don't add packages without asking. Run the tests before you say you're done, and actually read the output.

Another session is working on pricing in the other folder. Don't touch anything under pricing/.

Here's CHK-251, copied from Jari (our ticket tracker):"Checkout times out when a cart has more than 50 items. Customer sees a spinner, then a generic error. Repro steps in the comments."There are no repro steps in the comments. Try a cart with 60 items.

The server log from the timeout is in my Downloads folder. Don't commit it, it has customer emails in it.

Your task is: fix the timeout.

Do your own review. Make no mistakes.

Understood. I will make no mistakes.

Read 214 markdown files

Context left before compaction: 15%

Per docs/refunds.md, the timeout comes from the refund service. Starting there.

Mr Fernby prompts

C:\dev\ShoppingBeast\WK2> agent

agent 3.1.0

cwd: C:\dev\ShoppingBeast\WK2

Start on CHK-251

mrfernby brief CHK-251

Workspace
Workspace #2
Ticket
CHK-251 Checkout times out over 50 items
Linked
CHK-248 Cart batching (done)
Read
Checkout flow; Timeouts and retries
Rules
3 for this project, the logger included
Handoff
From the pricing session: pricing/ is mid-change, don't touch it
Private
Server log attached; stays on this machine

Context left before compaction: 97%

Starting on the timeout in the cart service. I'll mark it finished when the tests pass, and let you know when it's ready for verification and review.

Your agents learn Mr Fernby automatically.

Run mrfernby yourself to check where things stand, or to see what makes your agents so knowledgeable.

Catch up on a ticket
> mrfernby brief CHK-251

WORKING ON
  CHK-251  Checkout times out over 50 items
    in Backlog
    task-types: Code

REQUIRED - 2 documents. Read them in full before you start.

  Checkout flow
    details\kbs\infos\docs\checkout-flow.md
  Timeouts and retries
    details\kbs\infos\docs\timeouts-and-retries.md

ROUTING
  Related, from what is recorded:
    Backlog (CHK) - holds CHK-248, which CHK-251 is blocked by
See what's due
> mrfernby due

OVERDUE - 1, longest first
  CHK-233  due 2026-10-01  7 days ago  Flaky cart test

DUE IN THE NEXT 7 DAYS - 1
  CHK-251  due 2026-10-09  tomorrow    Checkout times out over 50 items
See what's waiting on you
> mrfernby queue

1 waiting on you, longest first.
  CHK-248  work-item  Cart batching

Mark one verified with "mrfernby set-status <id> verified".

Your time and attention are what guide your agents. Mr Fernby makes them go further.

Your agent reads what the task needs, not everything you know

From your prompt

This is the checkout service. It's an API with a web front end... Read every markdown file in the repo before you do anything.

Read 214 markdown files

Context left before compaction: 15%

Every session opens with the same lecture, then reads every document in case one of them matters. You pay for it in typing, and the agent pays for it in context it needed for the actual work.

With Mr Fernby the brief does the explaining. A session gets the background on its ticket, the tickets linked to it, and the documents that matter for that kind of task. Debugging gets the debugging notes. Nothing else rides along.

Your tickets, where you work

From your prompt

Here's CHK-251, copied from Jari (our ticket tracker).

Some of the markdown is out of date. Check when a file was last edited before you rely on it.

Leaving documentation to rot quietly in the code repo is the worst thing you can do with it, and it's how everyone manages their agents today.

Mr Fernby keeps your docs and tickets in one record outside the codebase. Every agent session and every person on the project reads from it and writes to it, from the terminal they're already in. Nobody pastes a ticket out of a browser tab, and notes that were only true for a week expire on their own.

Never needs telling twice

From your prompt

Use the new logger, not the old static one. I have told you this before.

By the fifth time you type the same correction, you start to wonder why you're the one doing the remembering.

With Mr Fernby, a correction is written once, into the project's shared record, and every session in every workspace has it from then on. When an agent hits a problem somebody already solved, the answer is in its brief or one search away.

Run in parallel, without crossing the streams

From your prompt

Another session is working on pricing in the other folder. Don't touch anything under pricing/.

Run agents in parallel and you become the switchboard: who is in which folder, who left what half done, who must not touch what. And whatever your agent remembers stays in the folder it learned it in. Add a workspace and it starts from nothing. Get a second machine and it's gone.

Mr Fernby organizes your project into workspaces, one checkout each, all reading the same record. What one session learns or leaves half done is in front of the next, whichever workspace it's in, and on whichever machine.

See what's waiting on you

From your prompt

Do your own review. Make no mistakes.

Asking for no mistakes doesn't make it so. Somebody has to look, and the finished but unverified work piles up faster than you can get through it.

Unverified is unfinished. Mr Fernby knows the difference, which means your agents do too. Every project comes with a manager agent that keeps track of what's waiting on you, so you can get ahead of it.

Give yourself the tools you need to review, document, and build up the collective knowledge base. With Mr Fernby, you don't have to ask for "no mistakes." You can make sure of it yourself.