Pre-built skills that give AI coding agents deep knowledge of the Next Commerce platform — APIs, CLI workflows, architecture patterns, and gotchas — so they can work autonomously on your store.
Skills are structured markdown files. Any AI tool that accepts a context file or system prompt can use them. They work with Claude Code, OpenAI Codex, Cursor, GitHub Copilot, and any other agent that reads markdown.
| Skill | What It Does | When to Use |
|---|---|---|
| Theme Figma Handoff | Prepare Figma storefront designs for implementation: source validation, section classification, asset manifests, Spark divergence ledger, matching visual refs, and documented coverage gaps | You have a Figma storefront/PDP/homepage design and need a low-inference handoff before next-theme-dev implements it |
| Theme Development | Build and customize storefront themes, including DTL templates, ntk CLI, Tailwind CSS, settings, side cart, active-theme publish safeguards, and visual QA | You're editing theme files, setting up a new storefront, publishing a bounded change, or debugging template and styling issues |
| Bulk Fulfillment Sync | Update orders to Fulfilled with tracking numbers from a CSV | Your fulfillment provider shipped orders but tracking didn't sync back — orders stuck in Processing |
| Bulk Fulfillment Move | Move fulfillment orders between warehouse locations in bulk — by order-number file or by Product ID / SKU list | Switching fulfillment providers, or moving every FO containing a given SKU/product to a new location |
| Bulk Subscription Actions | Pause, cancel, or PATCH explicit subscription fields for a list of subscription IDs | Merchant wants to bulk-pause until a date, set supplied renewal timestamps, bulk-cancel, or migrate subscriptions between gateways |
| Daily Ops Risk Scan | Read-only scan for Incomplete orders, Rejected orders, and Delivery Tracking failures/staleness | You want a daily queue of risky orders and stuck shipments to reduce support friction and dispute risk |
| New Campaign Setup | Scaffold and fully configure a new Next Commerce campaign repo — brand init, starter template, campaigns.json seed, API key, store details, and analytics in one pass | Starting a new Next Commerce campaign for a brand |
For design-led theme work, run next-theme-figma before next-theme-dev.
Both skills expect real screenshot evidence for visual QA, but they reuse tools
already available to the agent or operator and support manual capture. They do
not require a bundled browser automation package or screenshot service. When a
Theme Development task targets an active theme, the skill adds an exact file
manifest, rollback source, upload-count checks, and served-revision
verification.
Each skill directory contains two documents: a README.md — a plain-language
guide for the person running the skill (what it does, what you need, how to
ask for it), written for non-technical readers — and a SKILL.md, the
technical instructions the AI agent follows. Skill-specific detail lives in
those per-skill docs, not here. Related skills cross-reference each other in
their READMEs (for example, next-theme-figma runs upstream of
next-theme-dev, and next-campaigns-setup documents its boundary with the
campaigns-os package).
For most Next Commerce users, the most reliable path is to clone this repo and run the bundled installer. It previews changes before writing, supports common local agent profiles, and does not depend on external installer UX changing over time.
Clone and run the guided installer:
git clone https://github.com/NextCommerceCo/skills.git
cd skills
./skills.shPreview or install directly:
./skills.sh status
./skills.sh install codex
./skills.sh install codex next-ops-scan
./skills.sh status --target /tmp/next-skills next-ops-scan
./skills.sh dry-run --target /tmp/next-skills next-ops-scanstatus is read-only and prints both the source and installed SKILL.md
versions. A stale row means the installed semantic version is older;
modified means the versions match but files differ; local-newer prevents a
newer local copy from being mistaken for an ordinary refresh; and
unknown-version means one side does not use the required X.Y.Z format. Use
dry-run before install when reviewing several targets.
Targets:
claude->~/.claude/skillscodex->~/.codex/skillsagents->~/.agents/skillsall-> all of the above
Restart local agent sessions after updating skills so the refreshed instructions are loaded.
Updating an existing skill directory uses rsync so the destination path remains
present while files are refreshed. Install rsync before using skills.sh on
minimal environments that do not include it by default.
If you use an AI coding tool, you can also ask it to run the local guided installer for you:
Install the Next Commerce AI skill I need from https://github.com/NextCommerceCo/skills. Prefer cloning the repo and running
./skills.sh, choosing the installation location for my current AI tool. If a local checkout is not appropriate, use the publicnpx skillsinstaller or load the relevantSKILL.mdas context.
Tell it which skill you want, or ask it to inspect skills.json and choose the relevant one.
If you want a one-liner without keeping a local checkout, use the
skills CLI. It can pull SKILL.md
files from GitHub and install them into many agent-specific skill directories.
Install all skills globally for your detected agent:
npx skills add NextCommerceCo/skills -gInstall a single skill globally for Codex:
npx skills add NextCommerceCo/skills -g -a codex --skill next-theme-devList available skills without installing:
npx skills add NextCommerceCo/skills --listTarget a specific agent:
npx skills add NextCommerceCo/skills -g -a claude-codeSkip prompts for scripted installs:
npx skills add NextCommerceCo/skills -g -a codex --skill next-ops-scan -yUse npx skills update to refresh skills installed through the skills CLI.
Each skill is plain markdown. If your tool does not support native skill
directories or the installers above, load the relevant SKILL.md as a system
prompt, context file, rule, or chat upload according to that tool's conventions.
For AI agents that need to programmatically discover available skills, skills.json provides a structured manifest with skill IDs, descriptions, trigger phrases, and prerequisites. Agents can fetch this single file to decide which skill to load.
Each skill lists its own requirements in the file. Common across all skills:
- Access to a Next Commerce store
- An API key with the scopes specified by the skill (create at Dashboard > Settings > API Access)
Each skill is a directory containing a SKILL.md (the agent's technical
instructions) and a README.md (a plain-language guide for the person running
the skill). To add a new skill:
- Create a directory with a descriptive name (e.g.,
next-my-skill/) - Add a
SKILL.mdwith YAML frontmatter (name,version,description,allowed-tools) followed by the skill instructions in markdown - Add a "Using This Skill" section that points to the repo install guidance
- Add a
README.mdwritten for non-technical readers: what the skill does, what the person needs, and how to ask for it — plain language, no code examples (use tables for data examples and[!IMPORTANT]callouts for things that matter) - Add an entry to
skills.json - Update this README's skills table and any installer notes that should mention the skill
- Run
./skills.sh statusto confirm the local installer can discover the skill - Bump the skill version in both
SKILL.mdandskills.json - Run
python3 scripts/check_skill_versions.py --base origin/main - Open a PR
Every tracked file inside a skill directory is part of that skill's versioned
package. A PR that changes any file under next-*/ must advance that skill's
version in both its SKILL.md frontmatter and skills.json entry.
- Patch: fixes or clarifies existing behavior, safety rules, validation, or routing without adding a new execution surface.
- Minor: adds a new command, automation lane, detector/check family, output contract, or approved write surface.
- Major: changes ownership, safety posture, default side effects, or removes a supported contract.
CI compares each changed skill package with the pull request's base commit and rejects missing or non-increasing version bumps.
Work in this repo is tracked with GitHub Issues and coordinated on the
org-level Operations.
Kanban board (Todo / In Progress / Done). New issues are added to the board
automatically by the add-to-project workflow.
Before starting work on an issue: check it is not assigned to someone else,
assign yourself (gh issue edit <n> --add-assignee @me), and move the card to
In Progress. Open PRs with Closes #<n>; when the issue closes on merge, the
board's built-in "Item closed" automation moves the card to Done. Contributors
have a /next-board skill that wraps these board operations.