BRG Finance  ·  August 2026

Build Your Own
CoworkOS

Give Claude persistent memory, your finance context, and one governed process the whole team shares.

Finance · Governance Edition  ·  Claude Desktop App ·  Brought to you by Vivid
What a workstation actually is — and how it differs from a project Why Claude needs your context to be useful The three core files that power everything How the Governance layer keeps the team on one process Starter templates — copy and paste to get going Finance workstation ideas to start with
BRG Finance  ·  AI Enablement
BRG

Claude Tools — When to Use What

A quick reference for Finance team members across all roles

Browser  /  Mobile  /  Desktop
Claude Chat
claude.ai  ·  No setup required
Reach for it when
You need a quick answer, a sanity check, or a first draft — right now.
Good for
Quick research & summarisation
Paste a report or policy doc, ask what it means
Drafting on the fly
Emails, Slack updates, quick commentary summaries
Brainstorming & problem framing
Scenario thinking, stakeholder prep, slide structure
One-off file analysis
Upload a CSV or PDF and ask questions about it
Not ideal for
Work that needs your entity structure, FX conventions, or memory across sessions
Desktop App (Mac  /  Windows)
Claude Cowork
Installed locally  ·  Connects your tools
Reach for it when
You want Claude to know your entities, your numbers, and your reporting conventions — every session.
Good for
Recurring finance workflows
Variance commentary, board pack drafts, budget reviews
Working with local files
Read models, write narratives, output to Word or Excel
Persistent finance context
Entity structure, FX rates, budget versions — loaded automatically
Scheduled & repeating tasks
Monthly report snapshots, automated digests
Not ideal for
Writing or modifying code, running tests, or engineering workflows needing terminal access
Terminal  /  Command Line (CLI)
Claude Code
Developer tool  ·  Requires setup
Reach for it when
You're building automation, data pipelines, or custom tools that need to run code.
Good for
Finance automation scripts
Python for data extraction, consolidation, reconciliation
Data pipeline building
Connect ERP data sources, automate report generation
Excel/model scripting
VBA replacements, pandas transformations, SQL queries
Custom tool development
Build Finance-specific Cowork plugins and integrations
Not ideal for
Non-technical users, everyday comms tasks, or anything that doesn't involve writing or running actual code
The Problem

Claude starts cold every session. CoworkOS fixes that.

Without Cowork OS

Re-explaining context every time

Claude doesn't know your entity structure, fiscal year, budget version, or how you label variances. You spend the first few exchanges re-establishing basics before anything useful gets done.

Generic output. No memory of your chart of accounts, preferred rounding, or reporting conventions.

CoworkOS is just a folder on your computer. No data leaves Breville's environment.

A Fair Question

"OK, that sounds great. But what about all the project folders, skills and plugins I've already built in Claude Chat? Are they wasted?"

Be honest, a few of you are thinking it. Let's deal with it head on.

Why CoworkOS, Not Just a Folder

👀 CoworkOS vs a Cowork folder vs a Claude Chat project

Option A
Project in Claude Chat

Custom instructions plus a few uploaded docs, in the browser. Quick to set up and easy to share. But it can't open your local models, can't update its own memory, won't route tasks, and can't run on a schedule or hold the team to one process.

Best for: quick, shared, cloud Q&A.

Option B
Plain Cowork folder

Point Cowork at any folder and it reads your files, runs skills and connects your tools. Powerful, but with no CLAUDE.md or MEMORY.md it forgets between sessions, has no routing, and no governance — every analyst freelances their own way.

Best for: one-off local file work.

Your Chat projects aren't wasted — they're the rough draft. The instructions and context you wrote there lift straight into CoworkOS. Keep Chat for quick or shared questions; bring the monthly grind into CoworkOS, where it remembers, routes, and stays on-process.
The One Idea

A workstation is a specialist you brief once.

You explain the job properly one time — the entities, the FX conventions, the format, the sign-off chain. From then on you just ask for the thing, and it already knows. Like an analyst who's run the FX consolidation forty times and doesn't need the background again.

Does this job deserve its own workstation? Three questions
Does it come back?
Every month-end, every re-forecast, every board pack. Not once. If you'll only do it again in theory, it's not a workstation.
Do you re-explain yourself?
Same entities, same FX source, same "no, label it (U) not negative" correction. If you're typing the same context each cycle, that's the brief you should be saving.
Does "done" look the same each time?
Same columns, same variance format, same sign-off. If every output looks different, keep it in a normal chat.
Three yeses
Build a workstation. You'll get the hour back within a month — and for governed tasks, it locks in the approved process too.
Any no
Just ask Claude normally. No folder, no setup, no guilt.

Most of your work is not a workstation, and that's the point. They're for the handful of finance cycles you've already run a dozen times and will run a hundred more — budgets, board packs, FX consolidation.

The Question Everyone Asks
READ THIS ONE TWICE

The project is the context. The workstation is the do-er.

Four things that sound similar and aren't. Nine out of ten mix up the first two.

Claude Project
The context
In claude.ai  ·  a shared bucket
Think of it as
A folder of background reading you hand Claude before a conversation.
It holds
Reference documents
Policies, past board packs, source files
Standing instructions
Custom instructions for that topic
A chat history
Conversations grouped by subject
It does not
Touch your local models, follow a workflow, or run on a schedule. It waits to be asked.
Workstation
The do-er
In CoworkOS  ·  a folder that acts
Think of it as
A specialist you brief once, who then does the job on request.
It holds
A workflow
The steps, in order, every close
Its own rules and memory
What good looks like in this job — governed ones inherit the locked process
A place to put the output
Saved to your Outputs folder
It does not
Suit one-off work. If the job won't come back, you're doing setup for nothing.
Skill
The tool
Installed once  ·  used anywhere
Think of it as
A power tool on the bench. Any workstation can pick it up.
It gives Claude
One repeatable capability
Build an Excel model, write a board doc
Known-good method
Same technique whoever calls it — incl. finance-shared for governed tasks
Shareable packaging
Hand it to a colleague as one file
It does not
Know your entities, your files, or your standards. It's a capability, not a colleague.
MEMORY.md
The notebook
One per workstation  ·  plus a root one
Think of it as
What the specialist wrote down so you'd never have to say it twice.
It holds
Facts that change
FX rates used, budget versions, which entity
Decisions and why
So last month's assumption still stands
Corrections you've made
Say it once, it sticks
It does not
Hold rules. "Always label variance (F)/(U)" belongs in CLAUDE.md. Memory is facts, not instructions.
🔒 Governed workstations (Budget Modeller, FX & Consolidation, Board Pack Builder) must call the finance-shared skill for the Skill column above — they don't get to pick their own. See the Governance section for how that's enforced.
A Worked Example

Same month-end, twice. One version takes minutes.

Chat, or a project

You do the explaining, every cycle

  1. Find last month's variance file. Upload it.
  2. Explain which entities and FX rate source to use.
  3. Re-explain the materiality threshold you mentioned last time.
  4. Get commentary in the wrong format. Correct it.
  5. Copy the result somewhere you'll remember.
45 minand it resets next month

Not hypothetical. An AR specialist in APAC built three reconciliation workstations that now run in live monthly use. A finance manager in Canada built five.

How It Works

Four layers that power CoworkOS. One folder on your computer.

You  —  budget queries, variance questions, board pack requests
CoworkOS
CLAUDE.md + MEMORY.md
Entity context, fiscal conventions, routing map, and accumulated facts
voice-principles.md + Skills
Finance writing tone and installed capabilities
Governance
Agent
The traffic controller for finance work
For governed tasks, it mandates the approved finance-shared skill, plugin and model, then routes the work through the approved process. No DIY skills, no three different ways of doing budgets.
🔒 Set by Head of Finance
Work
stations
Governed · finance-shared process
Budget Modeller FX & Consolidation Board Pack Builder
Light-touch · your own way
Email Presentation + yours

The Governance Agent sits between your context and your workstations. Governed workstations (budgets, FX, consolidation) must use the shared, locked process so every analyst does it the same approved way. Light-touch workstations (email, slides) stay flexible. Each workstation is still one specialist in its own folder, but the governed ones inherit the finance-shared rules they can't override.

The Core Files

Three files power the whole system

File 01
CLAUDE.md

The global constitution. Tells Claude your entity structure, fiscal year conventions, number formats, and which workstation to route each finance task to. Every session starts here.

File 02
MEMORY.md

The persistent brain. Claude reads it at session start and writes to it when you say "remember this." Tracks active budget versions, FX rates used, reporting periods, and key decisions.

File 03
voice-principles.md

Your writing voice. How you write board commentary, CFO updates, and variance narratives. Claude reads this before drafting any finance comms or reporting on your behalf.

These three files live in your CoworkOS root alongside a 00_Resources/ folder. That's the whole system.
Setup Guide  ·  Three Phases

Let's get started.

You're one afternoon away from a Claude that knows your entities, your fiscal conventions, and how you write. Here's the path.

📁
Phase 01
Create your home base
Make an empty CoworkOS folder and run the Finance setup prompt — Claude scaffolds your entities, fiscal conventions, and routing in minutes.
🏗️
Phase 02
Build your workstations
Add a workstation for each recurring finance task — budget modelling, board packs, FX consolidation. Each gets its own rules and memory; the governed ones inherit the finance-shared locked process.
🔄
Phase 03
Keep it compounding
Log assumptions at the end of every session — FX rates, budget versions, rounding decisions. It builds your context and your audit trail automatically.
Phase 1 | Getting Started  ·  Prompt Template

Paste this to build your Finance CoworkOS from scratch

First-time setup

One prompt. Five minutes. Done.

Open Cowork with your CoworkOS folder selected, paste this prompt, and let it run. Claude creates all three core files with Finance-specific sections, including the two 🔒 LOCKED governance sections set by the Head of Finance, then asks a few questions to personalise the rest.

Before you paste: create an empty folder called CoworkOS somewhere on your computer (Documents is fine), then select it in the Cowork file picker.
I'm setting up CoworkOS for the first time.
I work in Finance at Breville Group. Please create:

1. CLAUDE.md — with sections:
   Memory System, Memory Routing, Finance Context,
   Preferences, Rules, Guardrails 🔒, Folder
   Structure, Routing Map, Change Control 🔒

2. MEMORY.md — with sections:
   About Me, Entity Context, Active Reporting
   Periods, Key Contacts, Key Decisions,
   Budget Versions

3. 00_Resources/voice-principles.md — with:
   Tone, Finance Writing Style, Sentence Style,
   Words to Avoid, Numbers, Words I Use

4. 00_Skills/     (empty folder)
5. 00_Scheduled/  (empty folder)

For Guardrails and Change Control, use the finance-
shared LOCKED defaults exactly as set by the Head of
Finance — don't invent or soften them. Mark both
🔒 LOCKED and note they ship via the finance-shared
plugin and a read-only master; a local copy can't
override them.

Use starter templates with [bracketed placeholders]
where I need to customise. After creating the files,
ask me 3 questions to personalise my setup for my
specific finance role and reporting responsibilities.
After Running the Setup Prompt

Let's see what you just created.

That one prompt created your entire CoworkOS skeleton — Finance-specific entity context, fiscal conventions, routing rules, and the two 🔒 LOCKED governance sections set by the Head of Finance.

The next three slides walk through each file with a copy-ready template. Fill in the [brackets] with your entities, FX conventions, and variance labels.

Once you've filled in your entity list, fiscal year, and reporting currency, add your first workstation. That's when Claude starts feeling like it actually knows your business.
📁 CoworkOS/
├── 📄 CLAUDE.md  — entity rules, routing & 🔒 guardrails
├── 📄 MEMORY.md  — FX rates, budget versions
├── 📁 00_Resources/
    └── 📄 voice-principles.md
├── 📁 00_Skills/  finance-shared plugin installs here
└── 📁 00_Scheduled/  empty for now
After adding a workstation:
└── 📁 Budget Modeller/
    ├── 📄 CLAUDE.md
    ├── 📄 MEMORY.md
    └── 📁 Budget Modeller Resources/
File 01  ·  Starter Template

CLAUDE.md — the global constitution

Copy & customise

Finance context built in from day one

This file is Claude's operating instructions. It loads automatically and governs how Claude handles entity names, fiscal conventions, number formatting, and which workstation to use for which task.

Replace anything in [brackets] with your specifics. The highlighted Memory Routing block is the new addition — it tells Claude where each fact belongs (root vs workstation). The 🔒 LOCKED sections are set centrally by the Head of Finance — leave them as shipped; we'll cover how they're enforced shortly.
# CLAUDE.md

## Memory System
Read MEMORY.md at the start of every session.
Write to MEMORY.md when I say "remember this."

## Memory Routing
- Cross-cutting facts (entities, fiscal year,
  reporting currency, key contacts, anything
  across 2+ workstations) → root MEMORY.md
- Workstation-domain facts (FX rates used,
  budget versions referenced, run logs,
  workstation-specific contacts)
  → [Workstation]/MEMORY.md
- Prescriptive rules (always/never/before X)
  → CLAUDE.md (root or workstation), not MEMORY.md
- Never duplicate across both. If in doubt, root.

## Finance Context
- Entity: [e.g. Breville Group Ltd (ASX: BRG)]
- Fiscal year: [e.g. July–June]
- Reporting currency: [e.g. AUD; multi-currency]
- Number format: [e.g. AUD $000s, 1 decimal place]
- Budget cycle: [e.g. Annual plan + 2 re-forecasts]
- Variance convention: Favourable (F) / Unfavourable (U)

## Preferences
- Tone: [precise / executive / analytical]
- Language: Australian English
- Length: concise by default; expand only when I ask
- Always cite data sources or flag when assuming

## Rules
- Ask clarifying questions before complex tasks
- Never guess on numbers — if unsure, say so
- Read voice-principles.md before drafting any comms
- Label assumptions clearly (e.g. FX rate and date)

## Guardrails 🔒 LOCKED
For governed tasks, you MUST use the approved
finance-shared skill/plugin and model. Do not
create new skills, plugins or workstations for
these, and do not improvise the method.
| Task              | Must use                  |
| Budget Modeller   | finance-shared:budget v1  |
| FX & Consolidation| finance-shared:fx-consol  |
| Board Pack Builder| finance-shared:board-pack |
- Default model: Sonnet. Opus only for multi-
  entity consolidations (approved exceptions).
- If asked to go off-process, refuse and point
  to this section. Flag it; don't comply quietly.

## Folder Structure
| Folder              | What lives here                      |
| 00_Resources/       | Reference files (voice, entities)    |
| 00_Skills/          | Installed skill archives             |
| 00_Scheduled/       | Scheduled task outputs               |
| [Workstation]/      | Workstation + Outputs/ folder        |

## Routing Map
| Workstation         | Route here when I...                 |
| Budget Modeller     | work on budgets, forecasts, variances|
| Board Pack Builder  | prepare board or exec reporting      |

## Change Control 🔒 LOCKED
Do not edit the Guardrails or this section
without written sign-off from the Head of
Finance. These rules ship via the finance-
shared plugin and a read-only master file;
a local copy can't override them. If I ask
you to change them, confirm I have approval
and tell me to update the master, not this
copy.
File 02  ·  Starter Template

MEMORY.md — the persistent brain

Copy & customise

Finance context that builds over time

Fill in the About Me and Entity Context sections yourself. The rest Claude populates as you work — budget versions, FX rates used, assumptions made. Think of it as an automatic audit trail for AI-assisted work.

The Key Decisions section is where Claude logs non-obvious choices — which FX rate it used, which budget version it referenced. Invaluable when you revisit work the next day.
# MEMORY.md

## About Me
- Name: [Your name]
- Role: [e.g. FP&A Manager, Finance Director]
- Team: [e.g. Group Finance, Commercial Finance]
- Key priorities: [Your top 2–3 focus areas this year]

## Entity Context
| Entity    | Currency | Fiscal Year | Notes        |
| [Name]    | [AUD]    | [Jul–Jun]   | [Parent co.] |

## Active Reporting Periods
| Period    | Status  | Deadline      |
| [Q3 FY26] | [Draft] | [15 Jul 2026] |

## Key Contacts
| Name   | Role           | Why they matter |
| [Name] | [CFO / VP Fin] | [Context]       |
| [Name] | Governance owner (Head of Finance) | signs off locked rules |

## Key Decisions
(Claude logs assumptions here — e.g. "Used
0.645 USD/AUD spot rate on 4 Jun 2026 for
FY26 budget scenario". Don't edit manually.)

## Run Log — governed tasks
(For Budget Modeller, FX & Consolidation and
Board Pack Builder, Claude logs the model and
skill version, key inputs, and date per run —
your audit trail. The LOCKED rules themselves
live in CLAUDE.md / the finance-shared plugin,
not here.)

## Budget Versions
(Claude tracks current versions — e.g. "FY26
Budget v2.1 approved 15 Mar 2026")
Memory Routing

Two places memories live. One simple rule.

When you say "remember this" inside a Finance workstation, Claude needs to know where to put it. Here's the routing decision — one test, every time.

Root MEMORY.md

Cross-cutting facts: your entities, fiscal year, reporting currency, key contacts, and anything that applies across two or more workstations. The shared brain for the whole Finance setup.

CLAUDE.md
(root or workstation)

Prescriptive rules — "always", "never", "before doing X do Y." These go in CLAUDE.md, not MEMORY.md. The 🔒 LOCKED governance sections already handle the most important ones.

The golden rule

Never duplicate the same fact across root and a workstation MEMORY.md. If it could plausibly live in both, default to root. One source of truth, no drift.

File 03  ·  Starter Template

voice-principles.md — your writing voice

Copy & customise

Finance comms that sound like you, not a template

Paste 5–8 of your own board commentaries, variance reports, or CFO updates and ask Claude to extract patterns. It'll populate your tone and style sections from your own writing.

Finance audiences hate hedging and vague language. "Revenue declined $2.3M (U) driven by lower APAC volumes" is better than "there were some revenue challenges." Specificity wins every time.
# Voice Principles

## Tone
- Sound like: [trusted adviser / sharp analyst]
- More formal with: [board / CFO / external]
- Avoid: [hedging / passive voice / imprecise language]

## Finance Writing Style
- Lead with the number, then explain it
- Variance direction: Favourable (F) / Unfavourable (U)
- Always state the period and comparison base
- Use "driven by" not "due to the fact that"
- Governed reports (board, FX, consolidation) follow
  the finance-shared LOCKED format; voice adds polish,
  not structure

## Sentence Style
- Short sentences for exec comms (20 words max)
- Longer for analyst-level detail
- Active voice: "Revenue grew 12% (F)"
  not "Revenue was seen to increase"

## Words to Avoid
- Never: "leverage," "synergise," "going forward"
- Avoid: "it is important to note that"
- No stacked adjectives or vague intensifiers

## Numbers
- Always include units ($, %, bps)
- Spell out millions: "$2.3M" not "2,300,000"
- Variances: always state direction (F) or (U)

## Words I Use
- [Your signature phrases — add them here]
- e.g. "driven by", "favourable mix", "on plan"
Governance  ·  Keeping the team on one process

How the Governance layer actually holds

Writing a rule in CLAUDE.md is a soft guardrail — Claude follows it, but a determined person can still edit the file. Real governance needs two halves: a clear policy, plus a structural lock that lives outside editable text.

Half 1 · The policy

Guardrails in the root CLAUDE.md

The Governance Agent layer, written down. It tells Claude what must be consistent and what stays free.

📌Mandates the approved skill, plugin and model for Budget Modeller, FX & Consolidation and Board Pack Builder.
🚦Routes each governed task through the finance-shared process — no DIY skills, no three ways of doing budgets.
🟢Leaves light-touch work free — email, slides and ad-hoc tasks aren't bound by finance-shared rules.
🛑If asked to go off-process, Claude refuses and flags it rather than quietly complying.
Half 2 · The lock 🔒

So staff can't quietly change it

The bit that makes "can't easily modify" real. The rules live where individuals can't rewrite them on a whim.

🧩Ship governed skills as one central plugin ("finance-shared"). Staff install it, they don't author it — editing a local copy doesn't change the distributed process.
🔒Hold the master CLAUDE.md read-only / centrally-synced (SharePoint or a managed repo). Only Finance leadership has write access.
✍️Change Control clause: locked sections need Head of Finance sign-off. Claude refuses to edit them in-session and points changes to the master.
Plus the operating controls that make it hold
📌
Versions pinned
Model and skill locked per task; re-checked on every upgrade
🧮
Tools do the maths
Numbers computed in a spreadsheet or skill, never the model's head
Human sign-off
A person approves before anything ships to the board or CFO
🧾
Run log
Model, skill version and inputs written to an audit trail
Phase 02  ·  Workstations

Time to build your first finance workstation.

Remember the one idea: a workstation is a specialist you brief once — Budget Modeller, Board Pack Builder, FX Consolidation. Pick the one task that costs you the most time every month. That's where you start.

Phase 2 | Workstations

Every workstation has the same four parts

A workstation is a specialist you brief once — and structurally, it's just a folder with three files. The four sections inside its CLAUDE.md define how Claude behaves when routing finance work there.

Resources

A table mapping reference files to "read when" conditions. Your chart of accounts loads only for modelling tasks; your board pack template loads only for reporting tasks. Context stays efficient.

Workflow

Numbered steps for the primary task. For a Budget Modeller: open data, identify variances, draft commentary, review assumptions, output. Start simple — 5–7 steps — and refine over time.

Editorial Rules

Always opens with "Follow voice-principles.md in 00_Resources." Then adds finance-specific rules — e.g. "Always label variance direction as (F) or (U). Always state the comparison period explicitly."

For governed workstations (Budget Modeller, FX & Consolidation, Board Pack Builder), these four parts sit on top of the finance-shared locked process — the workstation calls the approved skill and can't override it. Light-touch workstations (Email, Presentation) just use the four parts as they like.

Phase 2 | Workstations  ·  Prompt Template

Paste this to create a new Finance workstation

New workstation

30 seconds to a new capability

Pick one recurring finance task — budget variance commentary, board packs, FX consolidation. Fill in the two brackets, paste, and Claude scaffolds the whole workstation.

Good triggers are specific: "draft, review, or finalise any board or exec management reporting" rather than "reporting stuff." Specificity is what makes routing reliable.
Create a new workstation called "[Workstation Name]".

Route here when I: [describe the finance task — e.g.
"prepare, draft, or review any board pack or
executive management reporting"]

Create these items inside the workstation folder:

1. [Name]/CLAUDE.md with sections:
   - Identity (what routes here, what doesn't)
   - Resources (table: Resource | Read when...)
   - Workflow (numbered steps for the main task)
   - Editorial Rules (start with voice-principles.md
     reference; then add finance-specific rules,
     e.g. variance labelling, period references)

2. [Name]/MEMORY.md with:
   - Key Contacts (relevant to this domain)
   - Key Decisions (for this workstation)
   Note: cross-cutting facts (entities, fiscal
   conventions, key contacts) belong in root
   MEMORY.md, not here.

3. [Name]/[Name] Resources/  (empty folder)

If this is a governed task (budgets, FX,
consolidation), the Workflow MUST call the approved
finance-shared skill/plugin — don't create a new
one — and Identity should note it inherits the
LOCKED finance-shared process and can't override it.

After creating, add a new row to the Routing Map
in root CLAUDE.md.
A Quick Recap

One folder to rule them all. Four layers that power CoworkOS.

You  —  budget queries, variance questions, board pack requests
CoworkOS
CLAUDE.md + MEMORY.md
Entity context, fiscal conventions, routing map, and accumulated facts
voice-principles.md + Skills
Finance writing tone and installed capabilities
Governance
Agent
Mandates the approved skills, plugins and model
Routes governed finance tasks through the shared, locked process. Everyone does budgets, FX and consolidation the same approved way.
🔒 Set by Head of Finance
Work
stations
Governed · finance-shared
Budget Modeller FX & Consolidation Board Pack Builder
Light-touch · your own way
Email Presentation + yours

Global files apply everywhere. Each workstation is one specialist in one folder; the Governance Agent enforces the shared process on governed workstations, while light-touch ones stay flexible. The governed workstations layer on top but can't override the finance-shared rules.

Phase 03  ·  System Maintenance

Last bit. Log what you learn.

FX rates you used, budget versions you referenced, rounding decisions you made — none of it survives the session unless it's written down. One prompt fixes that, and builds your audit trail at the same time.

Phase 3 | System Maintenance

CoworkOS only gets smarter if you save what you learn

End of every session

Assumptions vanish when the session closes

FX rates you mentioned, budget versions Claude referenced, rounding decisions you made — none of it survives unless it's written to MEMORY.md. For finance work, that's also your audit trail.

One prompt at the end fixes this. Claude scans the whole conversation, then writes anything useful — assumptions, decisions, data points — to the right place.

Done consistently, this turns CoworkOS into a system that compounds. Every session makes the next one sharper, and every assumption is on record.
💬 Work
session
🔍 Scan
convo
📝 Write to
MEMORY.md
Smarter
next time
Scan this conversation for any unsaved assumptions,
decisions, data points, or useful context. Write
anything new to the right MEMORY.md:

- Cross-cutting facts (entities, key contacts,
  fiscal conventions, anything across 2+
  workstations) → root MEMORY.md
- Workstation-domain facts (FX rates used,
  budget versions, run logs, workstation
  contacts) → [Workstation]/MEMORY.md
- Prescriptive rules (always/never/before X)
  → CLAUDE.md, not MEMORY.md

For finance-specific items (FX rates, budget
versions, entity structures, rounding decisions),
log them under the appropriate section with the
date they were used.

Don't duplicate what's already there. Focus only
on what's genuinely new from this session.
Finance Workstations

Three workstations worth building first

🔒 Governed
WS 01
Budget Modeller

Route variance analysis, scenario planning, forecast updates, and budget commentary here. Give it your budget structure, comparison periods, and standard variance thresholds as reference files.

🔒 Governed
WS 02
Board Pack Builder

Route monthly and quarterly board reporting, exec update decks, and management commentary here. Teach it your standard pack structure, narrative style, and the CFO's preferred level of detail.

🔒 Governed
WS 03
FX & Consolidation

Route multi-currency reporting, intercompany elimination queries, and FX sensitivity analysis here. Give it your spot rates, hedging policy, and entity hierarchy as always-loaded reference files.

Not sure where to start? Board Pack Builder or Budget Modeller will cover 80% of recurring Finance time spend. All three run the finance-shared locked process, so every analyst's numbers come out the same approved way.
Get Started

What's the finance report you hate rebuilding every month?

That's your first workstation. CoworkOS compounds — the more you use it, the more Claude knows your models, your entities, and your voice. Start with one folder, three files.

Questions? Find us in #bai-claude-ama on Slack.

01
Create your CoworkOS folder
Make an empty folder in Documents. Select it in the Cowork file picker. That's your workspace.
02
Paste the first-time setup prompt
Slide 9 has the Finance prompt — hit Copy, open Cowork, paste, let it run. Three files and a folder structure in under a minute.
03
Fill in your Finance context
Add your entity list, fiscal year, reporting currency, and variance conventions. This is what makes every session feel like Claude already knows the business.
04
Add your first Finance workstation
Use the workstation prompt (Slide 17) to scaffold your first domain. Board Pack Builder or Budget Modeller are the natural first picks.
Efficiency

Keep your CoworkOS fast and cheap

Rule 01
Constrain the Root

Keep your master CLAUDE.md under 300 lines. Point to external reference files (entity lists, chart of accounts) rather than stuffing everything in one place — this keeps baseline token costs low.

Rule 02
Avoid Duplicated Rules

Never repeat global rules inside local workstations. If the root already says "label variances as (F) or (U)," your Budget Modeller shouldn't restate it. Inheritance is free.

Token efficiency means CoworkOS stays fast. A bloated root file taxes every session — trim it like you'd trim a good board pack.
Straight Talk  ·  Known Gaps

What this doesn't cover yet

Governance done well is friction plus deterrence, not a wall. Here's what it doesn't solve on its own, and what closes each gap.

Gap
It only binds inside CoworkOS
Plain Claude Chat, a phone, or a fresh folder skip every rule here. Close it: make the approved workstation the easiest path, and spot-check finance outputs that didn't come from it.
Gap
A locked rule still rots without an owner
"Head of Finance sign-off" needs teeth. Close it: name a governance owner, require two-person review on changes to the finance-shared skills, and review the lot quarterly.
Gap
Real numbers live on local laptops
CoworkOS folders hold sensitive figures and persist after people leave. Close it: add a device and offboarding step that wipes leavers' folders.
Gap
The data path is assumed, not confirmed
Pre-results and M&A data is the most sensitive we hold. Close it: confirm retention and enterprise terms with the agreement owner before sensitive data goes in.
Bottom line: treat this as a control that needs an owner and the occasional audit, not a one-and-done lock. The operating controls on the Governance slide handle the rest.
Phase 2 Consideration  ·  Shared Team Use

When Finance teams share a folder — or a Slack memory channel

The current model is personal by design. Shared folders and a Slack team memory channel add a fourth tier — and change where some facts should live.

The gap

The personal model breaks with shared folders

If the team shares one CoworkOS root, MEMORY.md becomes a jumble of personal context and team facts — two very different things in one file.

📌Root MEMORY.md is designed for one person. Your role, your preferences, your working context — not the team's agreed FX rate.
🚦Shared folders mix personal + team facts, making both unreliable. Personal context leaks to colleagues; team decisions get overwritten by individuals.
🟢Your Slack channel is already doing team memory. CoworkOS just doesn't have a formal relationship with it yet.
The clean pattern

Keep personal CoworkOS separate. Add a shared layer.

Three adjustments give you team memory without breaking the personal model or duplicating Slack.

🧩One CoworkOS per person. Don't share the root folder. Personal context stays personal — role, preferences, individual working notes.
📄Add a shared team-context.md in 00_Resources/ — owned by the governance owner, read-only for individuals. Holds agreed entity list, current budget versions, standing FX conventions.
✍️Slack = source of truth for team decisions. MEMORY.md stores a pointer, not a copy — e.g. "USD/AUD 0.645 — see #finance-ops 4 Jun 2026." No drift, no duplication.
The four memory tiers — where each fact type lives
👤
Personal MEMORY.md
Your role, preferences, contacts, individual working notes
🗂️
[Workstation]/MEMORY.md
FX rates used, budget versions referenced, workstation-specific notes
📄
team-context.md
Shared reference — entity list, standing conventions, governance-owner managed
💬
Slack channel
Team decisions, source of truth — MEMORY.md points here, doesn't copy
For the Head of Finance  ·  One-Pager

Governance at a glance, and what we need from you

One shared, approved way to do the finance work that has to be consistent — budgets, FX, consolidation — without slowing down everything else. Here's the shape of it and the handful of calls that are yours.

What's governed
Budget Modeller, FX & Consolidation, Board Pack Builder — all run the same approved finance-shared skill and model. No DIY methods.
What stays free
Email, slides and ad-hoc work — staff use Claude however suits them. Governance only touches what must be consistent.
How it's locked
Rules ship via a central finance-shared plugin and a read-only master file. A local copy can't override them; changes need your sign-off.
Decisions we need from you
1Approve the finance-shared process for budgets, FX and consolidation before it's locked.
2Name a governance owner and a backup who maintain the locked rules.
3Set the sign-off and review cadence — who approves changes, and how often we review.
4Confirm model pinning and the data path with the Anthropic enterprise owner.
Scheduled & Routine Tasks

Set finance routines to run on their own

Cowork can run a saved prompt on a schedule — daily, weekly, or at month-end. Build these once and they fire in the background, dropping dated outputs into your 00_Scheduled/ folder.

Daily FX rate log

Record the spot rates used each day to MEMORY.md, so every converted number has a dated, auditable source instead of a vague "around 0.65".

Weekly variance watch

Flag any actual-versus-budget variance over your threshold from the latest data drop, with a one-line likely cause where Claude can see one.

Quarterly governance review

Prompt the governance owner to review the locked finance-shared rules and skill versions. Keeps the guardrails current instead of quietly going stale.

Start with the one you do half-asleep every month. The month-end snapshot is the usual first win.
Reference

Glossary of Terms

Model The underlying AI engine. Claude comes in three sizes: Sonnet covers 80% of daily work; Opus is for complex, multi-step tasks with many interdependent decisions; Haiku is the fastest and cheapest for lightweight jobs. ⚡ Default to Sonnet. Reaching for Opus reflexively just burns tokens — and slows you down. Most Finance tasks don't need it. Don't be a #TokkenMaxxer. Think of the environment.
Agent An AI model that can take actions — read files, send messages, run tasks — rather than just answer questions. Claude in Cowork mode is an agent.
Workstation A specialist you brief once. A CoworkOS folder for one recurring finance job — Budget Modeller, Board Pack Builder — holding its own rules and memory. Claude routes work to the right one using your routing map; governed ones (🔒) also inherit the finance-shared locked process.
Governance Agent The layer that sits between your context and your workstations. It mandates the approved skill, plugin and model for governed finance tasks and routes them through the shared process.
Guardrails The 🔒 LOCKED rules in the root CLAUDE.md that say which approved skill/plugin/model each governed task must use. Set centrally; not edited by individuals.
Skill A packaged workflow that gives Claude a specialised capability — creating Excel models, writing Word docs, pulling data. Install once, trigger by name. Think of it as a power tool on the bench — any workstation can pick it up.
Session A single conversation with Claude. Context resets when it ends — which is exactly the problem MEMORY.md solves. Always log assumptions before closing.
MCP Model Context Protocol. The open standard that lets Claude connect to external tools and services — Slack, Jira, email, and so on. Think of it as Claude's plugin API.
Context Window The total amount of information Claude can hold in a single session. Larger context = more Claude knows, but also slower and pricier. Keep your root CLAUDE.md lean.
MCP Server A connector that implements MCP for one specific service. The Slack MCP server is what gives Claude read/write access to your Slack workspace.
Token The unit any LLM uses to measure text in and out. Longer prompts cost more tokens. The reason Sonnet is faster and cheaper than Opus for everyday Finance tasks.
Plugin A bundle of MCP servers and skills packaged together for easy installation. How capabilities are distributed and added to Cowork.
Routine A task set to run automatically on a schedule. "Every month-end, snapshot the reporting period status." Cowork runs these in the background without you having to ask each time.
Thank you
Questions? Find Vivid or Lilly on Slack.
#bai-claude-ama
1 / 29
← → to navigate