Community Skills Library

Purpose-built AI skills for accounting teams on Numeric. Each skill is a pre-built workflow your AI assistant can run on top of the Numeric MCP — close, flux, JE, recon, audit.

Starting points, not endpoints. Every skill here is battle-tested and gets you most of the way there. Customize them to your team's process by asking your AI assistant in any conversation — your modified version becomes the default.

How to think about these skills →

View on GitHub →

20Skills

7Categories

3Types

20Showing

All skills connect to Numeric via the Numeric MCP. Connect your workspace →

The new way of working

The keystrokes leave. The judgment stays.

These skills hand the rote work — pulling data, drafting, formatting — to your AI assistant, so your team spends its hours on the judgment only people bring: the review, the call, the relationships. See what AI-enabled close work looks like, and how the machine earns more of the routine over time.

Read the explainer →

Same work — fewer keystrokes, more judgment.

Gallery

Examples from the wild

End to End Accruals Processing using the Numeric MCP

Creating Journal Entries via the Numeric MCP

Skills

Browse the library

Highlight Essential skills QBO/Sage/Xero optimized

What it might cost to run your month-end close

Two core components: a fixed per-seat plan covering Claude across web, desktop, Code, and Cowork — plus variable token usage that scales with how often each workflow runs. Model the mix for your own team.

Open the cost calculator →

$200/mo

Plan — 10 seats at $20/seat, fixed

~$360/mo

Token usage — default close workflow mix, variable

~$56/seat/mo

Effective all-in, before any time saved

Illustrative — a 10-person team on the default workflow mix at June 2026 API rates. Swap in your own seat count, run volumes, and model choices in the calculator.

01

install

Start from a template.

A skill is a folder: a SKILL.md file that tells your AI assistant what to do, plus the scripts and prompts that do it. Download one from the library and install it in your tool of choice — no setup, no config screens. It activates automatically the next time you ask for something the skill knows how to handle.

Every skill here ships with sensible defaults written for a standard close. The community has already done the hard part of picking the right Numeric MCP calls, the right thresholds, the right output format.

02

invoke

Ask in plain language.

You don't pick a skill from a menu. You describe what you need — "run the accruals for April", "send overdue nudges for the close", "build me the rec workbook for prepaid" — and the right skill activates. Your assistant pulls the period and workspace context from Numeric, finds the matching task, and starts work.

Each run reads context from two places: the Numeric task description (period-specific notes from the preparer or reviewer) and the skill itself (your team's standing methodology). Together they give your assistant the full picture before any work begins.

03

refine

Tell it what to change.

A starter skill won't match your process out of the box. That's the point. When something is off — wrong format, wrong threshold, wrong channel — you say so in chat. Your assistant rewrites the skill in place.

"Update close-pulse to group overdue tasks by assignee, skip anything tagged 'on hold', and post the summary to #accounting-close in Slack instead of responding here."

Refinements can live in three places, depending on scope: the skill file for team-wide rules, the Numeric task description for period-specific notes ("exclude vendor AG Grid — bundled with the parent agreement"), or the close checklist for process-wide chaining. Your assistant picks the right place automatically.

04

ship

Your version is now the default.

The refined skill is just a file. Anyone on your team running the same skill picks up your rules automatically — no separate config, no shared spreadsheet, no tribal knowledge. Share it by dropping the .zip file into a teammate's chat, or commit it to git if you want full version history.

Skills can also self-track: after each run they log a comment back to the Numeric task — what was done, when, and the outcome. That creates a native audit trail and lets skills make smarter decisions next time (don't re-nudge someone nudged in the last 48 hours, don't re-post a JE that already cleared).

↻

evolve

Refine again, any time.

New entity, new approver, new materiality threshold, new reviewer DM channel — your process changes constantly. Tell your assistant. The skill keeps up. That's the difference between a skill and a static script: skills compound. Each round makes them more yours.

Skills are folders, not scripts Preferences persist across periods Shareable across your team

Contributing. These skills are built by the Numeric community. Built something useful? Ask your assistant to "bundle this into a skill", then send the file to support@numeric.io and we'll add it to the toolkit.