constraint-driven-development Skill for Claude: What It Does and How to Install It

Makes your agent write your project's quality bar into CONSTRAINTS.md and catch changes that quietly lower it.

Installs: 37.7K; on skills.sh
Stars: 112.6K; on GitHub
Files: 2; besides SKILL.md
Updated: Nov 2026; last repo commit

This skill, from the addyosmani/agent-skills pack, has the agent interview you with four questions and write a CONSTRAINTS.md file at the repo root. Each rule gets a number and a command that checks it. Every question has a default, so "I don't know" still produces a working setup.

It also installs the tools each dimension needs, wires the checks into the build, test, review and ship stages, and watches the diff for a weakened bar: new suppressions, skipped tests, stubs and lowered thresholds.

Good for

Developers letting agents write more code than they can read
Teams arguing about coverage or performance numbers in every pull request
Anyone whose agent keeps adding @ts-ignore or skipping tests to get to green
Maintainers setting up accessibility, security and performance gates

Install constraint-driven-development

Claude Code

Copy this prompt.

Install the constraint-driven-development skill in Claude Code. Run this command:

npx skills add https://github.com/addyosmani/agent-skills --skill constraint-driven-development

Then check that it loaded and tell me how to use it.

2
Paste it into Claude Code. It shows the change and asks before running it.

Requires Node.js.

From its listing on skills.sh. Source.

What's in the skill

When it loads

Establishes a project's quality bar as a written contract and stops agents quietly lowering it. Interviews the user on which dimensions matter, supplies sane default thresholds when they have no number in mind, records everything in CONSTRAINTS.md, and watches the diff for a weakened bar — new @ts-ignore or eslint-disable suppressions, skipped or deleted tests, assertions stripped out, unimplemented stubs, thresholds edited down. Use when no quality bar is written down, when the user says "set up constraints" or "define our standards", when the user wants dimensions they care about — accessibility, web performance, coverage — set up as enforced constraints, when an agent keeps silencing checks or skipping tests to get to green, when you need a coverage or performance threshold and don't know what number to pick, or when an agent writes more code than anyone will read.

Sections in SKILL.md

Overview
When to Use
Loading Constraints
The Process
Step 1: Detect before you ask
Step 2: Four questions, each with a default
Step 3: Write CONSTRAINTS.md
Floor (always enforced, no setup required)
Enforced with numbers
Measured, not yet enforced
Exceptions
Step 4: Install what each dimension needs
Step 5: Wire it to the lifecycle
Step 6: Guard the bar itself
Step 7: Ratchets, when you don't have a number
Sane Defaults

Key facts

Author: addyosmani (repo owner)
License: MIT
Installs on skills.sh: 37,706 (Nov 2, 2026)
GitHub stars: 112,555 for the whole addyosmani/agent-skills repo (Nov 2, 2026)
Last repo commit: Nov 1, 2026
What it includes: 3 files: SKILL.md, references/floor-guard.md (a reference check script) and scripts/quality-gate.mjs (a Node script for Semgrep and axe reports)
Needs: A live user for the interview. Each dimension you pick needs its own tool, such as Semgrep, gitleaks, osv-scanner, Lighthouse or axe
Works in: The README says the skills CLI installs into 70+ agents, including Claude Code, Cursor, Codex, Copilot and Cline

What it includes

Reads your stack, test runner, linters, CI and agent setup
Reports what it found in two lines before asking anything.
Try: “Set up constraints for this project.”
Read · Detect first
Which dimensions, block or warn, numbers, time budget
Each question has a guess and a default. It stops at four.
Read · Four questions
Floor, enforced rules, measured metrics, exceptions
Adds a line to AGENTS.md and CLAUDE.md telling agents to read it and not weaken it.
Write · Write CONSTRAINTS.md
Tools for types, lint, coverage, security, performance, accessibility
Examples: pip install mypy, pipx install semgrep, brew install gitleaks, npm i -D lighthouse. Adds check scripts to package.json.
Runs code · Install what each dimension needs
Fails the build on serious Semgrep or axe findings
Reads a report JSON and exits 0, 1 or 2. Copy it into your project's scripts/ folder.
Runs code · scripts/quality-gate.mjs
Fast checks per edit, heavier ones at task end and in CI
Maps checks to the build, test, review and ship stages with time budgets.
Write · Wire to the lifecycle
Spots five ways the bar gets lowered
Moved thresholds, easier tests, silenced checkers, unfinished stubs and new exceptions. references/floor-guard.md is a diff-based reference script.
Runs code · Guard the bar
Record today's number and refuse to get worse
Used when you have no target number.
Write · Ratchets

From the official docs

Permissions

Medium risk

The skill installs scanners and test tools, edits your config files and runs checks in your repo. Say which dimensions you want first and review the diff.

What it can do

Read your data
Create and change things
Delete things
Read-only mode

Watch out for

It installs tools on your machine
Step 4 lists commands such as pip install mypy, pipx install semgrep and brew install gitleaks. The skill says machine-wide tools can run in CI only if you prefer clean laptops. Say so before it starts. github.com
It edits your agent and project config files
It adds a line to AGENTS.md and CLAUDE.md and scripts to package.json. Review the diff. github.com
The interview needs a person present
SKILL.md says not to run it in CI, /loop or autonomous runs. There it applies only the floor and flags the rest for a human. github.com
gitleaks must use --redact
The skill says to report the rule and location, never the secret value, so leaked keys do not land in transcripts. github.com

Security advisories

No published security advisories as of Nov 2, 2026.

Technical details

Turning on read-only	No read-only mode. Use version control and review each change to CONSTRAINTS.md.
Talks to	Package downloads when tools are installed. Lighthouse and axe fetch the preview URL you give them. The Semgrep example uses the p/default registry rule set
Sign-in and keys	SKILL.md does not mention any account or key.
Scopes	Writes CONSTRAINTS.md and edits AGENTS.md, CLAUDE.md and package.json Installs tools with pip, pipx, brew or npm for the dimensions you pick Runs scanners and git diff commands in your repo; Lighthouse and axe need a running URL

Questions

What does the constraint-driven-development skill do?

It sets your project's quality bar as a written file with numbers and checks. It also watches the diff for an agent lowering that bar, for example with new suppressions, skipped tests or lowered thresholds.

How do I install it?

Use the install box at the top of this page. The README warns that a per-skill install copies only skills/<name>/, not the repo-level references/ folder. This skill's own references/floor-guard.md and scripts/quality-gate.mjs sit inside its folder.

Do I need to set up every check on day one?

No. SKILL.md says a first run can be floor-only, which is diff-based and needs no installs. You add dimensions such as coverage, security or accessibility as you install each tool.

What numbers does it suggest if I have none?

Its defaults include 80 percent coverage on changed lines, no dependency vulnerabilities at high or above, LCP of 2500 ms or less, CLS of 0.1 or less and zero critical or serious axe violations. If your project falls short, it suggests recording today's number and refusing to go lower.

Does it need the other skills in the pack?

Not for the steps it describes. SKILL.md says its interview borrows the one-question-at-a-time discipline from interview-me, and lists code-review-and-quality, ci-cd-and-automation, test-driven-development, security-and-hardening and performance-optimization under See Also. The Skill does not tell the agent to call any of them.

Is it safe?

It changes your project and installs software, so review each step. It edits config files (AGENTS.md, CLAUDE.md, package.json), installs scanners and test tools with pip, pipx, brew or npm, and runs scanners and git commands in your repo. Say which tools you want, review the diff, and note that the skill says gitleaks must use --redact so secrets stay out of output.

Which AI tools does it work in?

The README says the skills CLI installs into 70+ agents, including Claude Code, Cursor, Codex, Copilot and Cline. It also documents native plugins for some agents, which this page does not cover.

Information

updated Nov 2, 2026

Repository: addyosmani/agent-skills
Folder: skills/constraint-driven-development

addyosmani/agent-skills skills.sh

Sources

Docs, issue trackers and discussions this page is built from.

Show all 4

constraint-driven-development SKILL.md · github.com
addyosmani/agent-skills README · github.com
floor-guard.md · github.com
quality-gate.mjs · github.com