More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    Quality Bars That Hold: Constraint-Driven Development

    1

    Turn a quality bar into a written, checkable contract: a floor no one can waive, numbered rules with named checkers, enforcement at edit time and in CI, ratchets that raise the bar on evidence.

    $5

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    Quality Bars That Hold: Constraint-Driven Development

    Quality Bars That Hold: Constraint-Driven Development

    Example session with this skill installed

    I have a fictional habit-tracker app called Tidepool, built by a fictional solo developer. An agent writes most of the new code and the quality bar keeps slipping. Build me the contract: detect what exists, run the four-question interview with defaults where I defer, and give me the file plus the enforcement wiring.

    • Read your context and instructions
    • Compiled the quality bars that

    DETECTION — Tidepool, invented scenario and numbers for this demo only. Found: a typed language with a type checker, a test runner, a linter, CI running the full suite, and coverage on changed lines measured at 58%. No performance budget, no accessibility checker.

    INTERVIEW (defaults used where deferred): dimensions — types, coverage, secrets, security scanning. Block or warn — block the floor, warn on the rest for two weeks, then block. Targets — measure and hold. Tolerable slowness — 90 seconds at task end.

    CONTRACT — CONSTRAINTS.md at the repo root, last reviewed 2026-11-04, next review 2027-02-04. Floor: no new suppression comments without a documented exception; no stubs shipped; no skipped or deleted tests without a recorded reason; no secrets in source; the contract is never weakened to pass a change. Enforced: zero type errors at every edit; changed lines at least 58% covered at task end and in CI; no secrets in source at every edit; no high-severity security findings in CI. Deliberately not enforced: performance budgets (no staging URL to measure against), accessibility (no checker installed).

    ENFORCEMENT: edit-time checks are types, lint, and secret scan, blocking the edit; task-end checks are coverage on changed lines plus the quick suite, blocking completion until green or ruled on; CI runs the full suite, security scans, and blocks merge.

    FAULTS FOUND AND FIXED IN THIS RUN: the first draft set coverage at 80% from nowhere — an invented bar, rejected by the skill's own rule and replaced with the measured 58%. The first draft also omitted the deliberately-not-enforced section entirely; added with reasons before delivery. The quality gate passed all ten items after both fixes.

    CONTRACT — CONSTRAINTS.md

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Prose guidance like "write clean code" evaporates the moment a session ends, and an agent writes more in an afternoon than anyone will read that week. This skill moves judgment out of people's heads and into a written contract with numbers. It starts by detecting what the project already has — language and stack, test runner, linters, current coverage, CI, and the agent harness — and reports that in two lines, so it never asks what it can read. Then it runs a four-question interview, each with a default: which dimensions get enforced and what each costs, block or warn on failure, invented targets or measure-and-hold, and how slow a check may be. The contract itself is one file at the repo root: a non-negotiable floor covering suppression comments, stubs, skipped tests, secrets, and the rule that the contract can never be weakened to pass a change; a table of enforced rules with numbers, checkers, and run points; and a deliberately-not-enforced section that records what was rejected and why, so the same argument does not recur monthly. Enforcement is wired in three layers — fast checks at every edit, task-end checks inside a patience threshold, slow thorough checks in CI — and two ratchets keep the bar honest: measure-and-hold raises the line only on evidence, and a quarterly review asks whether each rule still earns its check time. The diff-watch list names the erosion patterns to review: suppression comments, skipped tests, and thresholds edited down.

    Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Code Quality & Review.

    • It produces a contract, not a test suite or linter configuration; the checks it names must exist or be added for the contract to mean anything.
    • Passing the contract does not mean the code is good; it means the code cleared the bar, and judgment still applies.
    • A contract nobody reviews decays like any other document; the quarterly review date is part of the file for that reason.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before buying a skill.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Money-back guarantee

    Doesn't do what its page promises? You get your money back.

    Trust & safety

    Security scan

    Security scan passed 3 days ago

    Permissions

    No special permissions declared or detected

    Listed3 days ago

    What's inside

    Frequently Asked Questions

    Yes. Detection says so plainly, and the contract's first rule becomes adding one, rather than inventing numbers for checks that do not exist.

    Hold it anyway. A measured line is defensible because it describes what the project already does, and the ratchet raises it on evidence as the number improves. An invented high bar gets ignored.

    That is exactly what the floor forbids and the diff-watch catches. Treat the weakening diff as a behavior change: rework the change, not the contract.

    New in Code Quality & Review