← All tools

Skills & plugins

Claude skills for operations teams

Map a process, write an AI agent’s job description, build a close checklist. The Baxter Mountain Skills Pack gives Claude the method our advisors use, so the output is structured, honest about what it does not know, and keeps a person on the decisions that matter.

Download the plugin Download single skills

Version 1.0.0Free, MIT licenceClaude apps and Claude CodeTry the previewHow to install

What’s in the pack

Three skills, three documents

Each skill is a short, tested method in a SKILL.md file. Claude loads it when your request matches. Install all three as one plugin, or pick the ones you need.

.01 · /agent-job-description

Agent Job Description

Turns a task or workflow into a job description for an agent, written like a role a business owner, IT and risk can sign off.

It produces

  • Role, inputs and allowed actions with limits
  • A never-do list
  • Human approval points and escalation routes
  • Metrics with baselines and an audit trail
agent-job-description-skill.zip

.02 · /close-checklist

Close Checklist Builder

Builds a month-end close checklist from a description of your entities, systems and close calendar.

It produces

  • Tasks by phase, each with one owner and a reviewer
  • Dependencies and business-day timing
  • Hygiene checks and the critical path
  • Markdown plus a CSV for your spreadsheet
close-checklist-builder-skill.zip

.03 · /map-process

Process Mapper

A quick first map of a process from notes or an SOP. For the full scored report, it points to Workflow Mapper.

It produces

  • A step table with owners and systems
  • Hand-offs, wait points and exceptions
  • A Mermaid flowchart
  • The top agent opportunity, with its approval point
process-mapper-skill.zip

Try it · preview

See a job description assemble

This is a preview of the skill’s output format, not the skill itself. It runs in your browser on simple rules and templates; nothing you type leaves the page. The real skill reads your actual workflow and writes every section from it.

1 · Pick a sample task
Why this matters
An agent’s scope comes from the task, not the technology. Describe the work a person does today, in one or two sentences, and name the team. Vague tasks produce vague permissions, and vague permissions are where risk hides.
4 · What does the work touch?
Why this matters
Each of these adds a guardrail: a line on the never-do list and a point where a person approves. Watch the approval count change as you tick them.
6 · Sign-off checks

Preview of output format · Sample data

Email me this
0Human approval points
0Never-do rules
0Allowed actions

Ready to sign off?

    What this means

    Write one from your real workflow with the skill

    Install

    Set it up in a couple of minutes

    Use the skill zips in claude.ai and the Claude apps, or the skills or plugin in Claude Code. The method is the same everywhere.

    1. .01

      Download the skills

      Get one zip per skill: agent-job-description, close-checklist-builder and process-mapper. Keep them zipped.

    2. .02

      Upload them in Settings

      In Claude on the web at claude.ai or in the desktop app, open Settings → Capabilities, find Skills and upload each zip. Skills need code execution switched on in the same settings; on Team and Enterprise plans an owner may need to enable skills first.

    3. .03

      Ask in plain language

      Paste your notes or attach a document.

      Write a job description for an agent that does this task:
[paste the task or workflow here]

    Field guide

    Three things to know before you install

    What a skill is, how to scope an agent like a hire, and how to keep a close checklist honest. Useful with or without the pack.

    .01

    A skill is a method, written down

    A skill is a folder with a SKILL.md file: a short description that tells Claude when to use it, then the steps, an output template and examples. Claude reads the description all the time and the full method only when your request matches, so skills cost little until they are needed. Think of it as the checklist an experienced colleague would hand you.

    Rule of thumbIf you paste the same instructions into Claude more than twice, they belong in a skill.
    .02

    Write an agent’s job like a new hire’s

    Name the outcome it owns and the person accountable for it. List what it reads, what it may write, and what it must never do. Then mark where a person approves and who it calls when it is unsure. In our experience, most arguments about agent risk end once this page exists, because everyone can see the boundaries.

    Rule of thumbEvery write action needs a log entry, and every action that moves money, changes access or speaks for the company needs a person.
    .03

    Checklist hygiene beats checklist length

    A close checklist fails quietly: tasks owned by “Finance”, reviews done by the preparer, a dependency that runs backwards. One owner per task, a reviewer who is not the preparer, and timing in business days catch most of it. Then look at who carries the busiest day.

    Rule of thumbIf one person owns more than about a third of the tasks on any day, that day sets your close date.
    Myth

    Guardrails slow an agent down, so start with broad access and tighten later.

    Reality

    Narrow access and clear approval points are what let a team say yes to an agent at all. It is far easier to widen a scope that has a clean audit trail than to explain an action nobody approved.

    Field note

    When we scope a first agent with a finance team, the never-do list is usually written faster than the allowed actions. People already know what must not happen. Writing it down first gives risk and IT something concrete to approve.

    Questions

    What is the difference between the plugin and the skill zips?

    The plugin is for Claude Code. It installs all three skills plus slash commands in one step. The skill zips are one skill each, for uploading in claude.ai and the Claude apps, or for unzipping into your Claude Code skills folder.

    Is the “try it” preview the skill?

    No. The preview shows the skill’s output format, assembled in your browser from simple rules and templates. Nothing you type leaves the page. The skill reads your real workflow, asks questions where needed and writes each section from your source.

    Will the skills make up numbers?

    No. Each skill is told never to invent volumes, thresholds, baselines or deadlines. Missing facts are marked “Not stated” and listed as open questions, and suggested thresholds are labelled “Proposed” for the owner to confirm.

    How does Process Mapper relate to Workflow Mapper?

    Process Mapper is the quick version: one table, one diagram, one opportunity. Workflow Mapper is the full method with scored opportunities and a detailed first agent. If both are installed, Process Mapper hands off to Workflow Mapper when you want the full report.

    Do the skills work with our systems?

    The skills work from what you give them: notes, SOPs, exports or descriptions. They do not connect to your ERP or other systems. Name your systems in the request and the output will use those names.

    Can you help us put an agent into production?

    Yes. Scoping the job, building the agent with your team and keeping people in the loop on key decisions is the core of our work.

    Stay in the loop

    Get updates and new skills

    We turn methods from our client work into skills. Leave your email and we will send new ones as they ship.

    Get updates and new skills

    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