How to use the Accounting Software Selection agent skill

What is an agent skill, and what does this one make the agent do?

An agent skill is a folder with a SKILL.md file and supporting reference files. When it is installed, the agent reads these instructions when a request matches the skill's description, and follows them.

This skill asks for your shortlist and business requirements, then records what each vendor documented or demonstrated. It keeps facts separate from evaluator opinions. It scores shortlisted packages against 57 evidence-backed fields, and can produce a short recommendation or, on request, CSV, SQL DDL, JSON Schema or Notion-style outputs built from one field list.

• It does not pick a winner, announce a best option, or record an approval. The business makes the final decision.
• Capabilities nobody has verified are marked Untested, not as a missing feature.
• Nepal tax and statutory claims (VAT, PAN, TDS, IRD reporting, e-billing, CBMS, payroll, SSF) stay Untested until current evidence from the vendor or authority is held.

How do you install the skill?

Download the ZIP file for this skill from this page. Then use one of the two options below.

For Claude Code, unzip the file into ~/.claude/skills/ to make it available across your projects, or into a project's .claude/skills/ folder to limit it to that project.

In the Claude apps, open Settings, go to Capabilities, and upload the skill under Skills.

• Personal install: ~/.claude/skills/
• Project install: .claude/skills/ inside your project folder
• Claude apps: Settings → Capabilities → Skills

What does the skill need before it runs?

The skill lists no required tools or prerequisites. It does not run scripts and does not use the network. Everything it does happens in the conversation, using the information you give it.

• Requires: nothing listed
• Runs scripts: no
• Uses network: no

How do you trigger the skill with a natural request?

You do not need a special command. Describe what you need in plain words. The skill is triggered by requests about choosing or comparing accounting software, and by spreadsheets or documents that already hold a scored evaluation.

The agent reads your request and works out what you want. If you want to compare packages, it asks for the shortlist first. If you already have a spreadsheet, it can reorganise it. If you only ask how to do something, it answers directly.

• Compare accounting software for a trading business in Nepal with these three packages.
• Set up a vendor demo evaluation sheet for our shortlist.
• Turn our current evaluation spreadsheet into a consistent scoring record.
• Build a three year software cost comparison from these quotations.
• Give me the CSV for this evaluation.

What questions will the agent ask, and in what order?

The agent asks one short question at a time and skips anything you have already answered. It starts with the shortlist, then the type of business, scale, requirements and their priority, Nepal requirements, platform needs, money, and finally the output you want.

Requirements are ranked as Must-have, Should-have, Nice-to-have or Not required, so a failed Must-have can be flagged as a deal-breaker. Prices are taken only from a written quotation.

What outputs can the skill produce?

By default you get a short recommendation rather than a spreadsheet. Artifacts are produced only when you ask. All formats are built from one field list, so they do not drift apart.

Each row keeps evaluation status, selection decision, rejection reason and deal-breaker flag as separate fields. The three Rating fields hold the named evaluator's judgement; capability fields hold what the vendor documented or demonstrated.

• CSV
• SQL DDL
• JSON Schema
• Notion mapping (uses the notion-manual-import skill if installed; otherwise a manual mapping is given and no live workspace is changed)

What are the safety limits?

The skill does not run scripts and does not use the network, so it cannot pull prices or vendor data from the internet. It only uses what you give it or what a named source gives it.

It is not a bookkeeping, tax-filing, vendor contracting or legal tool. It may review evaluation summaries you share, but it should not ask for unnecessary personal or confidential business records.

• Do not paste sensitive personal or confidential records you do not need to share.
• Treat Nepal tax and compliance statements as needing current evidence from the vendor or authority.
• Always check the final decision yourself; the skill is designed not to make it.

Tips for better results

Give the shortlist and the Must-have requirements early. The more precisely you rank requirements, the more useful the deal-breaker flags become.

Keep evaluator names with their ratings. An average across evaluators hides disagreement, so the skill keeps each assessment separate.

• Bring a written quotation if you want cost figures in the record.
• Use the same demo tests for every candidate so the results are comparable.
• Say which currency you are working in.