product-context-builder — Claude Code skill: features and setup
The product-context-builder Claude Code skill builds or revises files that give AI agents product and repository context. It runs when a project needs agent instructions, product principles, design guidance, or task-specific skills.
MIT ★ 107 updated 2026-11-08
npx skills add thedesignproject/agent-skills --skill product-context-builder -a claude-codeWhat product-context-builder does
The skill inspects a repository, then uses its existing material to draft or update AGENTS.md, CLAUDE.md, PRODUCT.md, DESIGN.md, and project skills. It checks current instructions, build and test commands, UI components, design tokens, and product documentation. Files are drafted one at a time for review.
It chooses files based on the project rather than generating the whole set by default. A final audit checks each result against repository evidence and lists decisions that need human confirmation. Commands, component names, product positioning, and constraints must come from the project; the skill does not invent them.
What it helps with
- assemble agent instructions from repository rules
- record product principles in
PRODUCT.md - document an evidenced design system in
DESIGN.md - separate shared repository guidance from task-specific workflows
- find conflicts among project context files
Example prompts for the agent
- Prepare repository context for AI agents and inspect the existing guidance.
- Draft
PRODUCT.mdfrom the project materials and list decisions that need confirmation. - Check whether the project has enough evidence for a
DESIGN.md.
How to install product-context-builder
1Skills CLI (Vercel)
npx skills add thedesignproject/agent-skills --skill product-context-builder -a claude-code2Manually: put the files in place
~/.claude/skills/product-context-builder/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/product-context-builder/
1Skills CLI (Vercel)
npx skills add thedesignproject/agent-skills --skill product-context-builder -a codex2Manually: put the files in place
.agents/skills/product-context-builder/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add thedesignproject/agent-skills --skill product-context-builder -a cursor2Manually: put the files in place
.cursor/skills/product-context-builder/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install thedesignproject/agent-skills skills/product-context-builder2Manually: put the files in place
.github/skills/product-context-builder/SKILL.md1Skills CLI (Vercel)
npx skills add thedesignproject/agent-skills --skill product-context-builder -a gemini-cli2Platform CLI
gemini skills install https://github.com/thedesignproject/agent-skills --path skills/product-context-builderManually: put the files in place
.devin/skills/product-context-builder/SKILL.mdThe legacy .windsurf/skills/ path still works
Skills CLI (Vercel)
npx skills add thedesignproject/agent-skills --skill product-context-builder -a opencodeClaude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about product-context-builder
How do I install product-context-builder in Claude Code?
Run npx skills add thedesignproject/agent-skills --skill product-context-builder -a claude-code. The skill file is at ~/.claude/skills/product-context-builder/SKILL.md.
What does product-context-builder do?
It examines a repository and helps create agent instructions, product guidance, design documentation, and project skills. It also checks the results against project evidence and flags decisions for human review.
Which files can product-context-builder create?
It works with AGENTS.md, CLAUDE.md, PRODUCT.md, DESIGN.md, and project skills. The set depends on the task; it does not generate every file automatically.
Source text
SKILL.md · MIT
--- name: product-context-builder description: Create or improve a repo's AGENTS.md, CLAUDE.md, PRODUCT.md, DESIGN.md, and project skills so AI agents can use the product and design system accurately. --- # Product Context Builder Build a compact, connected context system grounded in the user's real product and repository. The finished files should help an AI agent understand the codebase, the product judgment behind it, the visual system it must reuse, and the task-specific workflows it can load when relevant. ## Start by discovering what exists Inspect the repository before interviewing the user. Locate: - existing `AGENTS.md`, `CLAUDE.md`, `PRODUCT.md`, `DESIGN.md`, and skill folders - package scripts, build and test commands, source directories, and contribution rules - UI components, Storybook stories, theme files, CSS variables, Tailwind configuration, and token packages - product documentation, PRDs, research, brand guidance, and accessibility requirements Report what you found, what can be inferred safely, and what needs a human answer. Never invent commands, component names, product positioning, tokens, or constraints. ## Choose the files that earn their place Do not generate every file automatically. - Create or update `AGENTS.md` when multiple coding agents need stable repository instructions. - Create or update `CLAUDE.md` when Claude Code needs tool-specific guidance that is meaningfully different. Keep shared rules aligned with `AGENTS.md`. - Create `PRODUCT.md` when agents need durable product judgment: audience, purpose, personality, principles, references, and anti-references. - Create `DESIGN.md` when a real visual system or component library can be documented from evidence. - Create a project skill only for a repeatable task with a recognizable trigger and a stable workflow. Read the matching guide before drafting a file: - [Always-loaded files](references/always-loaded-files.md) - [PRODUCT.md](references/product.md) - [DESIGN.md](references/design.md) - [Project skills](references/project-skills.md) - [Validation](references/validation.md) Use the templates in `assets/templates/` as prompts for evidence, not forms that must be filled mechanically. Remove empty and irrelevant sections. ## Work in reviewable passes Draft one file at a time and explain: 1. which repository evidence shaped it 2. which statements came from the user 3. which questions or conflicts remain Ask the user to review strategic judgment before treating it as durable context. Product personality, anti-references, exceptions, and hard constraints need explicit confirmation. ## Keep the layers connected The always-loaded file should contain the smallest useful routing rule. For example: ```md Before writing or changing UI, read PRODUCT.md and DESIGN.md. Reuse documented components and tokens. If the system does not cover a design need, ask before creating a new pattern. ``` Do not copy the full contents of `PRODUCT.md` or `DESIGN.md` into `AGENTS.md` or `CLAUDE.md`. Do not hide universal repository rules inside a task skill. ## Finish with a context audit Read [validation](references/validation.md), check every generated file against the repository, and summarize: - files created or updated - evidence used - contradictions resolved - assumptions still awaiting confirmation - the next real task the user should test with the new context system
© thedesignproject · MIT · thedesignproject/agent-skills