Install and Use Guide — Factory by NewEcom.AI
Factory by NewEcom.AI Get the plugin

Install & use guide

Install and use the Factory skills

Five free, calibrated AI-assisted delivery skills for Claude Code, Cursor, Codex CLI, and Gemini CLI. Install once, then trigger each skill in plain language across the delivery loop — from wireframe to sprint review.

/plugin marketplace add NewEcomAI/factory-skills-public /plugin install factory@factory-skills

Works in Claude Code, Cursor, Codex CLI, and Gemini CLI. No account required.

1 · Install

Install the plugin

The five skills are authored once as skills/<name>/SKILL.md (the Agent Skills open standard) and work across four coding agents. The install step differs per platform.

Claude Code

In a Claude Code session (the interactive terminal), add the marketplace, then install the plugin:

  1. /plugin marketplace add NewEcomAI/factory-skills-public
  2. /plugin install factory@factory-skills

The part after @ is the marketplace name (factory-skills), not the repo path. Set scope with --scope user (default, all projects), project (writes .claude/settings.json to share with your team), or local. Requires Claude Code 1.0.33+.

Cursor

Cursor reads skills/<name>/SKILL.md directly — no marketplace step. Clone the repo into your workspace or copy skills/ into your project and the skills load as Agent Skills.

Codex CLI

Codex CLI also reads SKILL.md natively — point it at the skills/ directory. To tune the CLI listing, copy platform/codex/<name>/openai.yaml to skills/<name>/agents/openai.yaml.

Gemini CLI

Gemini CLI loads an extension: place the platform/gemini/ folder under .gemini/extensions/factory-skills/ (with the repo available so the @{skills/<name>/SKILL.md} references resolve). Each skill is exposed as a /<name> command.

2 · How it works

Trigger skills in plain language

Each skill is triggered by natural language — describe the task and the agent loads the matching skill. There are no commands to memorize. For example:

  • “Turn this wireframe into a backlog with estimates.”
  • “Draft a change request for this new scope.”
  • “Build me a sprint review deck from these metrics.”

Every skill is cold-install safe: with no project config, the skill asks you inline for what it needs (product name, stack, infrastructure areas, response window, sender name). You can start with zero setup.

3 · Workflow

Where each skill fits

  1. Kickoff — brand-spec-generator to lock visual identity, then wireframe-to-backlog to turn mockups into an estimated, sprint-planned backlog.
  2. Per story — story-to-prompt to convert each backlog item into a precise, guard-railed prompt before handing it to a coding agent.
  3. Mid-project — change-request-impact when scope shifts, to size the change and generate the client communication.
  4. End of sprint — sprint-review to produce the client-facing deck.

4 · Skill reference

Five skills, end to end

wireframe-to-backlog

Wireframe → estimated backlog

When: you have a wireframe, mockup, or Figma/HTML prototype and need a realistic, estimated plan.

What it does: a 5-pass read that produces epics and stories with hour estimates and sprint assignments, applying a 4× correction factor to infrastructure stories (auth, payments, billing, webhooks) — the work routinely under-estimated.

Output: a structured backlog (epics → stories → hours → sprint plan) ready to paste into your tracker. Review the discovery-gap section for always-required stories (CI/CD, monitoring, access control).

brand-spec-generator

Brief → design-system spec

When: you need a consistent, accessible visual language before or during build.

What it does: generates a 13-section design system with color tokens and a type scale, each checked for WCAG 2.1 AA contrast, plus a ready-to-paste .cursorrules block.

Output: a full HTML design-system spec and a cursor rules export. PM tip: the “NOT-list” is the highest-leverage input.

story-to-prompt

Story → agent-ready prompt

When: before handing any backlog story to a coding agent.

What it does: produces a structured prompt with context, invariants, explicit tasks, validation gates, acceptance criteria, and a mandatory file allowlist. One portable body, tuned per platform (Cursor, Claude Code, Codex CLI, Gemini CLI).

Output: a copy-paste implementation prompt. PM tip: keep the file allowlist tight so the agent only touches what it should.

change-request-impact

Scope change → impact + email

When: a scope change arrives mid-project and you need to respond quickly and defensibly.

What it does: classifies the change, sizes the effort, and runs an absorbability test against your remaining sprint capacity.

Output: an internal assessment record and a client-facing change-request email stating the impact, the options, and the response window.

sprint-review

Metrics → 10-slide deck

When: at the end of a sprint, to produce a polished client review.

What it does: renders a fixed 10-slide structure — cover, KPIs, metrics, shipped, story table, cadence, quality, conventions, handoff, closing.

Output: a .pptx deck. Populate DATA{}, then run npm i pptxgenjs and node generators/sprint_review_gen.js.

Free vs calibrated

What the free tier includes

The free skills ship a fixed calibration — the 4× infrastructure factor and a static estimation snapshot — enough to produce realistic plans and reviews out of the box.

Calibrating estimates to your team’s live delivery data, and auto-populating the sprint-review DATA{} from your GitHub / Linear activity, is a Factory MCP feature. The free tier always works without it.

Troubleshooting

If something doesn’t work

  • The skill didn’t activate. Be explicit about the task (“turn this into a backlog”), or invoke it by name where slash commands are supported.
  • It’s asking questions I expected it to know. That’s the cold-install path — answer inline once, or supply a project config later to skip the prompts.
  • The sprint-review generator errors. Ensure npm i pptxgenjs has run and DATA{} is populated; it writes <Product>_Sprint<N>_Review.pptx to the current directory.
  • Install fails on Claude Code. Add the marketplace first, and remember the install target is factory@factory-skills (the marketplace name), not the repo path.

Ready to build?

See the five skills → Have us run the build →