Skills & plugins

Turn process notes into a workflow map

Workflow Mapper is a Claude skill and Claude Code plugin. Give it interview notes, an SOP or a process document and it returns a clean map: steps, owners, systems, hand-offs, wait points, exceptions and a ranked list of where an AI agent could take on work first.

Download the plugin Download the skill

Version 1.0.0Free, MIT licenceClaude apps and Claude CodeTry the live demo How to install

Interview notes · 10 notes

Workflow map · 5 mapped steps

Step 1Invoice arrives

Step 2Agent matches PO

Hand-offAP to approver

Bottleneck3-day approval wait

Step 3Treasury pays

Map locked

Map ready: 4 steps mapped, 1 bottleneck uncovered

Sample data

Map ready: 4 steps mapped, 1 bottleneck uncovered

Sample data

Try it

Map a workflow in ten seconds

Pick a workflow or type your own. The page sketches the kind of map the plugin produces: steps by owner, hand-offs, waits, exceptions, where an agent fits and where a person must decide. It runs in your browser on built-in patterns. The plugin works from your real notes.

Why map one workflow at a time?

An agent takes on a job inside one process, with one trigger and one end state. If you map “finance”, you get a fog. If you map “invoice exceptions”, you can see who waits on whom, which step has clear rules and where a person has to stay in charge.

Workflow map

Invoice approval

The plugin writes the same Mermaid swimlanes into its report. Paste into GitHub, Notion or mermaid.live to render.

What this means

Agent opportunities, ranked Each factor scored 1 to 5, out of 25

Why these five factors?

An agent does well where work is frequent, the rules can be written down, the data sits in systems it can read, mistakes are cheap to catch and you can measure the result. A step that takes the longest is not always the best candidate. A step that waits on a person often is.

Map your real notes with the plugin

The full report

From interview notes to a working map

Below is a sample: notes from a short interview about invoice exceptions, and the map the skill produced from them. Hover or tab through the steps to see what it pulled out.

Input · interview notes Sample

Interview with the AP lead, accounts payable, 12 March. Topic: what happens when an invoice does not match.

Invoices come into the ap@ shared inbox, mostly as PDFs. About 1,200 a month.
An AP specialist keys each one into NetSuite every morning.
NetSuite runs a three-way match against the PO and the goods receipt. Roughly 1 in 5 fail, usually a price variance or a missing receipt.
When one fails, the specialist puts the bill on hold and emails the buyer in Procurement.
Buyers take two to three days to reply. If the receipt is missing, they chase the warehouse to post it.
Variances over $500 or 2% need the AP manager to approve. Under that, the specialist can adjust.
Sometimes the vendor has to send a credit memo. That can take a week.
Once it is sorted, the specialist releases the hold.
Payments only go out in the Thursday run, after the controller approves it.
“Everything is tracked in a shared spreadsheet. Nobody trusts the status column.”

Output · sample workflow map

Hover or focus a step to see its owner, system, wait and flags.

Sample step table
#	Step	Owner	System	Wait / volume	Flags

Bottlenecks

Buyer replies take two to three days. Every failed match waits on an email answer, and about one invoice in five fails.
A weekly payment run. A hold released on Friday waits up to six days for the Thursday run.
Credit memos take about a week and sit outside any system the team can see.
Status lives in a spreadsheet. Nobody has one trusted view of which exceptions are open or why.

flowchart LR
  subgraph AP["AP specialist"]
    S1(["Receive invoice"]) --> S2["Key into NetSuite"]
    S4["Email buyer, hold bill"]
    S8["Release hold"]
  end
  subgraph ERP["NetSuite"]
    S3{"Three-way match?"}
    S10(["Pay vendor"])
  end
  subgraph PROC["Procurement buyer"]
    S5["Check PO and receipt"]
  end
  subgraph VEND["Vendor"]
    S6["Send credit memo"]
  end
  subgraph FIN["Finance approvers"]
    S7{"Approve variance?"}
    S9["Approve Thursday run"]
  end
  S2 --> S3
  S3 -- "Match" --> S8
  S3 -- "No match ~20%" --> S4
  S4 --> S5
  S5 -- "Resolved" --> S8
  S5 -- "Credit needed" --> S6
  S5 -- "Over $500 or 2%" --> S7
  S6 --> S8
  S7 --> S8
  S8 --> S9
  S9 --> S10
  classDef wait stroke-dasharray: 5 5;
  class S5,S6,S9 wait;

Paste into any Mermaid renderer, such as GitHub, Notion or mermaid.live, to get the same swimlanes.

Agent opportunities, ranked

Each candidate is scored 1 to 5 on five factors, out of 25. Sample output from the notes above. The top pick drafts and chases; a person stays in the loop for every decision that moves money.

Sample agent opportunity scores
Rank	Opportunity	Steps	Volume	Clear rules	System access	Contained risk	Measurable	Total

Suggested first agent

Invoice exception agent

Ranked first, and the lowest-risk place to start: it drafts and chases, while people keep every decision that moves money.

Role: When a match fails, classify the exception, pull the PO, receipt and vendor history, draft the buyer query with the evidence attached, follow up after 24 hours and keep the exception log current.
Permissions: Read NetSuite bills, POs and receipts. Send internal email from the AP inbox. Write to the exception log. Never release a hold or change an amount.
Human approval: The AP specialist approves every adjustment. Variances over $500 or 2% still go to the AP manager.
Metric: Median exception cycle time. Baseline not stated in the notes, so measure it for two weeks before go-live.

Field note

The first map a team draws is the official process. The second, drawn after talking to the people who do the work, is the real one. Notice the last line of the notes above: nobody trusts the status column. That line is not in any SOP, and it points straight at an agent opportunity.

What's inside

A method, not just a prompt

The skill carries the same approach our advisors use in discovery: find the hand-offs and the waits first, then score where an agent can safely take on work.

SKILL.md
The mapping method
Extract steps, owners and systems, then find hand-offs, wait points, rework loops and exceptions. Never invents volumes or durations.
Mermaid flowchart
Swimlanes by owner
A flowchart LR with one subgraph per owner, decisions, exception labels and wait points marked.
Scoring
Ranked agent opportunities
Volume, clear rules, system access, contained risk and measurable outcome, each scored 1 to 5.
First agent
A job description for the agent
Role, permissions, the human approval point and the metric to track from day one.
/map-workflow
A slash command
In Claude Code, point it at a file or paste notes: /map-workflow docs/ap-sop.md.

Install

Set it up in a couple of minutes

Use the skill zip in the Claude apps, or the skill or plugin in Claude Code. The method is the same everywhere.

.01
Download the skill
Get workflow-mapper-skill.zip. Keep it zipped.
.02
Upload it in Settings
In Claude on the web or desktop, open Settings → Capabilities, find Skills and upload the zip. Skills need code execution switched on in the same settings; on Team and Enterprise plans an owner may need to enable skills first.

.03

Ask for a map

Paste your notes or attach a document and ask in plain language.

Map this process and show me where an agent could help:
[paste your interview notes or SOP here]

Option A · personal skill

Copy the skill folder into ~/.claude/skills

Download the skill zip, then from the folder you saved it in:

mkdir -p ~/.claude/skills
unzip workflow-mapper-skill.zip -d ~/.claude/skills/

This creates ~/.claude/skills/workflow-mapper/SKILL.md. Start a new session and ask Claude to map a process.

Option B · plugin

Add the plugin, with its slash command

Download the plugin zip. It is laid out as a small marketplace. Unzip it:

unzip baxter-workflow-mapper-plugin.zip -d baxter-workflow-mapper-plugin

Then, inside Claude Code, from the same folder:

/plugin marketplace add ./baxter-workflow-mapper-plugin
/plugin install baxter-workflow-mapper@baxter-mountain

Run it on a file or on pasted notes:

/map-workflow docs/invoice-exceptions-sop.md

Field guide

Three things a workflow map teaches you

What our advisors look for first when a map lands on the table. Use these with the plugin, or with a whiteboard and a marker.

.01

The cost lives between the boxes

Most steps take minutes. The days go to work sitting in someone’s inbox after a hand-off, or waiting for a weekly batch. Count the lane changes and the wait points before you look at any single step.

Rule of thumbIf a step only exists to pass work to someone else, it is a candidate for an agent or for deletion.

.02

Exceptions are where the hours hide

The happy path is often already automated. People spend their time on the cases that do not fit: the mismatch, the missing document, the unusual request. An agent that triages and chases exceptions, with the evidence attached, tends to pay back faster than one that speeds up the normal path.

Rule of thumbAsk “what happens when it goes wrong, and how often?” for every decision on the map.

.03

Mark the human checkpoints first

Before you add any agent, mark the steps where a person must decide: anything that moves money, changes access or speaks for the company. Agents prepare the work around those points. In our view, most tasks should stay at or below about 80% automated, with people owning the decisions.

Rule of thumbIf an error would be costly or hard to undo, an agent drafts and a person approves.

Myth

The best first agent goes on the step that takes the longest.

Reality

The best first agent goes where work waits between people, the rules are clear and a person still approves the outcome. Long steps are often long because they need judgment.

Questions

What can I give it?

Interview notes, meeting transcripts, SOPs, policy documents or a rough description typed from memory. It works best with one process at a time, described by the people who run it.

Will it make up numbers?

No. The skill is told never to invent volumes, durations or costs. Anything your source does not state is marked “Not stated” and listed as an open question, so you know what to ask next.

Is the demo on this page live?

Partly. The “Try it” mapper runs in your browser on built-in patterns and simple keyword rules; nothing you type leaves the page. The full report is a pre-written sample. The plugin itself reads your actual notes, so your results will depend on your source material and the Claude model you use.

Can you map our processes with us?

Yes. Mapping is the first step of our discovery work. We interview the people who own the work, map the process with them and turn the top opportunity into an agent in production.

Stay in the loop

Get updates and new skills

We are building more skills from our client work, including an agent job-description writer and a close checklist builder. Leave your email and we will send them as they ship.

Get updates and new skills Book a discovery call

Want a tool built for your team?

We build custom assessments, skills and internal apps as part of our engagements. Tell us what you need.

Book a discovery call leron@baxtermountain.ai