Module 4 Intermediate 45 minutes
4 objectives
  • Build a customized CLAUDE.md that gives Cowork persistent context about your role and tools
  • Know which connectors are available and which need workarounds
  • Install at least one plugin for your role, with awareness of the security tradeoffs
  • Combine multiple data sources into a single output, and understand where matching breaks down
📁 9 files Start Workshop →

Connecting Cowork to Your Tools — Jira, Slack, Confluence, HubSpot — live data instead of CSV exports

1Hook5 min
2Challenge15 min
3Setup12 min
4Challenge8 min
5Learn5 min
1

The Before/After Demo

5 min

The Before/After Demo

5 min
CLAUDE.md improves output relevance by giving Cowork persistent context about your role, tools, and terminology. It is a low-effort configuration step — about 10 minutes to set up. The payoff scales with how often you use Cowork: more sessions means more cumulative time saved on prompt editing.

Emma in Compliance spends 20 minutes every prompt adding the same context: "I use our KYC portal, our SLA is 48 hours, we call it EDD not enhanced checks, the MLRO signs off on high-risk..." Every. Single. Time. What if Cowork remembered?

Same prompt. Two different outputs. The difference: a tiny text file called CLAUDE.md.

WITHOUT CLAUDE.md: "Draft a checklist for reviewing a high-risk merchant."
Result: Generic checklist. Mentions "check financial statements" and "review sanctions lists." No reference to your AML screening tool, MLRO, or your actual SLA. You would spend 20-30 minutes rewriting it to be useful.

WITH CLAUDE.md: Same prompt.
Result: References your actual tools (your KYC portal, your AML screening tool, your identity verification tool). Uses your terminology (EDD, PEP, SAR). Includes your SLA (48-hour escalation). Closer to usable, though still needs a review pass.

The difference is a short text file that tells Cowork about your context. Here is how to build one.

📋 2

Build Your CLAUDE.md

⏱ 90015 min
CHALLENGE⏱ 15:00

The Task

Pick the template matching your role from the data pack. Open it. Replace every [PLACEHOLDER] with your real information — your tools, your terminology, your SLAs, your rules.

Then save it as CLAUDE.md in your Cowork project root folder.

Test it: Ask Cowork the same question from the demo ("Draft a checklist for reviewing a high-risk merchant" or equivalent for your role). Compare the output quality before vs. after. The improvement is usually noticeable on the first try, but CLAUDE.md is something you refine over time as you notice gaps.

Data pack file: claude-md-templates/

Choose Your Prompt

After saving your CLAUDE.md, test it: Draft a weekly chargeback case review checklist for cases due this week.
After saving your CLAUDE.md, test it: Prepare a call brief for a £200K prospect in the restaurant sector.
After saving your CLAUDE.md, test it: Draft a checklist for reviewing a high-risk merchant flagged by your AML screening tool.
After saving your CLAUDE.md, test it: Create a summary of my team's key processes and tools.

Scoring

Self-assessment: rate output quality 1-5 before CLAUDE.md vs 1-5 after. The delta is your score.

Verification Checklist

0/6 verified
🔧 3

Connect and Query

12 min

Connect and Query

12 min
1

Open the Cowork connectors panel. Check what MCP connectors are already installed. You may see: Jira, Confluence (these are the verified ones at our company). As of March 2026, Slack and DocuSign connectors are also available from Anthropic — but availability depends on your admin's configuration.

2

If Jira is connected: try a real query — 'Show me all open Jira issues assigned to me in my current project.' If Jira is NOT connected: use the data pack file jira-export-sample.csv instead — ask Cowork to 'Read jira-export-sample.csv and tell me which issues are blocked and what is blocking them.'

3

Open the Plugin Marketplace (or slash command list). Find one plugin relevant to your role. Install it. Try one slash command from the plugin.

36% of marketplace skills have been found to contain security issues (per community audits). Skills lack versioning and signing. Stick to Anthropic-created plugins unless you can review the source. If a skill requests access to files or tools that seem unrelated to its stated purpose, do not install it.
4

Note which connectors are available, which are not, and which tools you will need CSV export workarounds for. This is your current integration map. For most teams, your helpdesk and your CRM are not directly connected — you will need to export CSVs from those tools and bring them into Cowork manually.

5

If your company already uses an enterprise AI platform (Glean, Microsoft Copilot, or similar): those are 'ask a question' tools — great at searching company knowledge and internal docs. Cowork is a 'build something' tool — it creates files, processes data, and runs on a schedule. They're complementary, not competing. Note in your integration map which tasks belong to each.

📋 4

Multi-Source Fusion

⏱ 4808 min
CHALLENGE⏱ 8:00

The Task

You know those mornings where you have a spreadsheet open in one window, your issue tracker in another, and your risk dashboard in a third — and you are manually cross-referencing between them? That is what this exercise automates.

Cross-referencing multiple data sources is one of Cowork's stronger capabilities. It handles the mechanical matching; you verify the output.

This is the most common pain point in FinTech: copying data between systems that don't talk to each other. Compliance copying from their screening tool to the ticketing system. Chargebacks copying from scheme portals to spreadsheets. Sales copying from CRM to reports. The Cowork approach: export both sources as CSV, let Cowork do the matching, verify the output. Not magic — but faster than alt-tabbing between four browser tabs.

Important caveat: fuzzy matching (e.g., merchant names that do not match exactly between files) is unreliable. Cowork works best when files share an exact key like Merchant_ID. If your files use slightly different name formats, expect missed matches and false joins. Always check the unmatched rows.

Data pack file: jira-export-sample.csv + merchant-risk-profiles.csv

Choose Your Prompt

I have two files in my workspace: jira-export-sample.csv (open issues) and merchant-risk-profiles.csv (merchant risk data). Play detective: (1) Cross-reference them — for each issue mentioning a merchant, pull in their risk level and alert count. (2) Flag any HIGH or CRITICAL risk merchants that ALSO have open issues — these are your "double trouble" cases that need immediate attention. (3) Identify merchants with risk profiles but NO open issues — are they being monitored? (4) Create an Excel output sorted by risk, with a separate "Attention Required" sheet for the double-trouble cases. Add a 3-line narrative summary: how many merchants are flagged, what the biggest risk pattern is, and your recommended next step.
I have two files in my workspace: jira-export-sample.csv (Jira issues) and merchant-risk-profiles.csv (merchant risk data). Cross-reference them: for each Jira issue that mentions a merchant name, look up that merchant's risk level from the risk profiles. Create an Excel output with columns: Issue_Key, Summary, Status, Priority, Merchant_Name (if found), Risk_Level (from profiles), Alert_Count. Flag any HIGH or CRITICAL risk merchants with open Jira issues. Sort by Risk_Level descending.

Scoring

All 5 checks pass = Gold. 3-4 = Silver. 1-2 = Bronze.

Verification Checklist

0/5 verified
5

Ship It

5 min

Ship It

5 min

Take stock of what you built in this module:

  1. CLAUDE.md — a persistent context file. It makes future prompts more relevant by giving Cowork your role, tools, and terminology. You will refine it over time as you notice gaps in output quality.
  2. A connector/plugin inventory — you know what is wired up directly and what needs CSV export workarounds. This is a practical constraint map, not a wishlist.
  3. A multi-source output — a cross-referenced file that demonstrates Cowork's ability to join data. Useful when the join keys are exact; unreliable when they are not.

Assess honestly: did the CLAUDE.md make a noticeable difference in your test prompt? If not, your template may need more specific information — tool names, SLAs, terminology. A vague CLAUDE.md produces vague improvements.

Steal This Prompt

Copy these prompts, customize the file paths to match your workspace, and use them in Cowork. Replace ~ with your actual home directory path if needed.

Cross-Reference Jira + Risk Profiles

ComplianceOperationsEveryone
Read jira-export-sample.csv and merchant-risk-profiles.csv from my workspace. Cross-reference: for each Jira issue menti...
Show full prompt
Read jira-export-sample.csv and merchant-risk-profiles.csv from my workspace. Cross-reference: for each Jira issue mentioning a merchant, add the merchant's Risk_Level and Alert_Count from the risk profiles. Create an Excel output sorted by Risk_Level descending. Flag HIGH/CRITICAL merchants with open issues.

Jira Board Summary

Everyone
Read jira-export-sample.csv. Group issues by Status and Priority. Show: total count, blocked issues (with what's blockin...
Show full prompt
Read jira-export-sample.csv. Group issues by Status and Priority. Show: total count, blocked issues (with what's blocking them), critical/high priority items needing attention. Format as a scannable briefing.

"The Pre-Meeting Briefing"

LeadershipProductEveryone
I have a meeting in 30 minutes. Below are: (1) the agenda, (2) notes from the last meeting on this topic, and (3) any re...
Show full prompt
I have a meeting in 30 minutes. Below are: (1) the agenda, (2) notes from the last meeting on this topic, and (3) any relevant data or context. Create a 1-page pre-meeting briefing with: key context I need to remember, decisions that need to be made in this meeting, suggested positions on each decision with brief reasoning, questions I should ask, and any risks or concerns to raise. Keep it scannable — I'll read this in the lift on the way to the meeting room.

[Paste agenda, previous notes, and context here]

"The Decision Log"

OperationsComplianceEveryone
I'm pasting a chat thread where our team discussed a process change. It got messy. Extract the following: (1) What was u...
Show full prompt
I'm pasting a chat thread where our team discussed a process change. It got messy. Extract the following: (1) What was ultimately decided — the final outcome, not the debate. (2) Who made the final call (or if it was a group consensus). (3) What alternatives were considered and why they were rejected. (4) Any caveats or conditions attached to the decision. (5) Next steps and who's responsible. Format as a clean "Decision Record" document with date, participants, and a clear structure. Save as decision-log.md.

[Paste the chat thread here]

"Meeting Notes to Ticket Drafts"

Everyone
I'm pasting notes from a meeting. Extract three things: (1) **Decisions made** — what was agreed, stated as facts. (2) ...
Show full prompt
I'm pasting notes from a meeting. Extract three things:

(1) **Decisions made** — what was agreed, stated as facts.
(2) **Action items** — who owns what, with deadlines if mentioned.
(3) **Draft Jira tickets** — for each action item, draft a ticket with: Title (imperative verb + what), Description (context from the meeting + acceptance criteria), Priority (suggest based on urgency language used), and Assignee (from the action item owner).

Format decisions and action items as a summary at the top. List draft tickets below, ready to copy-paste into your issue tracker.

[Paste your meeting notes here]

"The CDD Follow-Up Email Drafter"

ComplianceOperations
I have merchants with incomplete CDD/KYC documentation. For each merchant below, I'll list: name, what documents are mis...
Show full prompt
I have merchants with incomplete CDD/KYC documentation. For each merchant below, I'll list: name, what documents are missing, and how many days overdue.

For each merchant, draft a follow-up email that: (1) addresses them by name, (2) explains which documents are still needed in plain language (not compliance jargon), (3) states the consequence of not submitting (e.g., settlements remain on hold), (4) provides a clear deadline, (5) keeps a professional but firm tone — this is a compliance requirement, not a request. Save as a Word document, one email per page.

[Paste your merchant list: name, missing docs, days overdue]

"The Policy Translator"

ComplianceHREveryone
I'm pasting a policy document (compliance policy, HR policy, security policy, or similar). It's dense and nobody on my t...
Show full prompt
I'm pasting a policy document (compliance policy, HR policy, security policy, or similar). It's dense and nobody on my team will read the whole thing.

Create two outputs: (1) A **1-page plain-English summary** — what the policy requires in simple language. No jargon. Written for someone who needs to follow the rules but doesn't need to know the legal reasoning. (2) A **Quick Reference Card** — a bullet-point checklist of the 5-10 things my team actually needs to DO differently because of this policy. Format as: "DO: [action]" and "DON'T: [action]."

Flag anything in the policy that's ambiguous or could be interpreted multiple ways — mark it [ASK COMPLIANCE].

[Paste the policy document here]

"The Vendor Comparison Matrix"

OperationsLeadershipEveryone
I'm evaluating tools/vendors for [purpose]. Here's what I know about each option: [Paste info about each vendor: name, ...
Show full prompt
I'm evaluating tools/vendors for [purpose]. Here's what I know about each option:

[Paste info about each vendor: name, pricing, features, pros, cons, notes from demos]

Create an Excel workbook with: (1) Sheet 1 — Comparison matrix: rows = criteria (price, features, integrations, support, security, scalability), columns = vendors. Score each 1-5 with a brief justification. (2) Sheet 2 — Pros/cons summary per vendor in plain English. (3) Sheet 3 — Recommendation: rank the vendors with reasoning. Flag any criteria where I don't have enough info to score — mark as [NEED MORE INFO].

Be honest — if two vendors are basically the same, say so.

Key Takeaways

  1. CLAUDE.md is a 10-minute setup that improves output relevance on every future prompt. It is not a one-time task — update it as your tools and processes change.

  2. Jira and Confluence connectors work via MCP. Your helpdesk and your CRM likely need CSV export workarounds for now — this is a platform limitation, not a configuration issue.

  3. Multi-source fusion (combining 2+ data sources) works well when files share exact keys. Fuzzy name matching is unreliable — verify join results manually.

  4. Vet any community plugins carefully. 36% have been found to contain security issues. Stick to Anthropic-created ones unless you can read the source.

Safety Note:

CLAUDE.md security: Never store credentials, API keys, or tokens in your CLAUDE.md file. It is a context file, not a secrets vault. Cowork reads it on every session start — anything in that file is exposed to the model.

Plugin security: 36% of marketplace skills have been found to contain security issues (per community audits). Skills lack versioning and signing. There is no sandboxing between plugins and your workspace files. Stick to Anthropic-created plugins unless you can review the source code. If a plugin requests permissions that seem unrelated to its stated purpose, do not install it.

Chrome automation: Computer Use (mouse/keyboard control, added Mar 23) takes screenshots of your screen. Never use it on banking portals, healthcare systems, HR tools with sensitive employee data, or any system containing PII. Cowork is a research preview — Anthropic states it is not suitable for regulated workloads.

Connector data: Data pulled through MCP connectors (Jira, Confluence, etc.) passes through Cowork's context window. Be mindful of what you query — avoid pulling sensitive customer data, PII, or restricted information into Cowork sessions.

Materials

Download these files to use with the exercises above. Previews load automatically.

📊
jira-sprint-export.xlsx20 Jira issues with priority highlighting for connector exercises
Download
📊
merchant-risk-profiles.xlsx15 merchant risk profiles with risk-level colour coding
Download
📕
compliance-policy-extract.pdfSample AML/KYC policy extract (PDF reference)
Download
📝
claude-md-templates/chargebacks-analyst.mdCLAUDE.md template for Chargebacks
Download
📝
claude-md-templates/sales-rep.mdCLAUDE.md template for Sales
Download
📝
claude-md-templates/compliance-analyst.mdCLAUDE.md template for Compliance
Download
📝
claude-md-templates/customer-support.mdCLAUDE.md template for Support
Download
📝
claude-md-templates/finance-analyst.mdCLAUDE.md template for Finance
Download
📝
claude-md-templates/generic.mdGeneric CLAUDE.md template
Download