claudecodex

ratchet-loop

Ratchet Loop — a checkpoint-and-gate build loop for AI coding agents, working with any model provider: a single orchestrator splits a feature into small ordered checkpoints, and every checkpoint must pass four hard gates (behavior tests, UI review against a prototype, adversarial code review, human review) before the next one starts. Includes the checkpoint planner, test planner, prototype/design agents, isolated UI reviewers, adversary + fixer, a Stop hook that won't let the agent quit early, and a checkpoint viewer.

4 on GitHubby Rahil Piraniv1.14.0Apache-2.0updated 12 days ago

Not in a marketplace yet, so it loads for one session with --plugin-dir.

Plugin ratchet-loop in rahilp/Ratchet-Loop. Commands read from the repo's manifest — 12 days ago.

This plugin runs code on your machine — its hooks fire automatically on events like tool use. The exact commands are listed below; read them before installing.

What it installs

Skills (6)

  • ratchet-adversarial-loop

    Runs the Ratchet adversarial code-review gate — independent reviewers (another company's model through any registered agent CLI whenever one is available and working, otherwise the host's company on a different model, otherwise the host's own sub-agents) assume the checkpoint's code is wrong and hun

  • ratchet-design-md

    Creates or updates DESIGN.md — a structured, machine-readable design-system file (tokens for color, type, spacing, radius, elevation, motion; component specs with states; layout and content rules) extracted from an existing codebase or built from a brief. Prototypes, implementers and UI reviewers al

  • ratchet-orchestrator

    Runs the Ratchet build loop (a checkpoint-and-gate workflow) end to end — splits a feature, screen, migration or refactor into small ordered checkpoints, gets the human to approve the plan once, then builds each checkpoint through four hard gates (behavior tests, UI review against a prototype/refere

  • ratchet-prototype

    Builds a single-file, clickable HTML prototype of a feature that follows the project's DESIGN.md, with every UI state switchable by URL (?state=…) and stable data-testid hooks, so it can be demoed to a human or client and used as the pixel reference for automated UI review. Use for "prototype", "moc

  • ratchet-tdd-planner

    Plans the behavior tests for one Ratchet checkpoint before any code exists — turns "done when" criteria into concrete, ID-tagged test cases (happy path, edge cases, permissions, failure modes), decides which can be proven in code without a browser and which must go to the UI gate, and writes tests/C

  • ratchet-ui-compare

    Runs the Ratchet UI gate — captures the reference (HTML prototype, old app screen, or design export) and the real app in the SAME state, replays the same interaction journey on both, then has two isolated reviewers (visual fidelity + interaction behaviour) that never see project instructions list ev

Subagents (13)

  • ratchet-adversary

    Ratchet adversarial reviewer. Assumes the checkpoint's code is wrong and tries to prove it — hunting for violations of the project's written standards (lens=standards) and for bugs, edge cases, security and data-integrity problems (lens=correctness), or both. Every finding cites a rule or a concrete

  • ratchet-app-scout

    Ratchet loop scout. Checks the current state of the app before a build — detects the stack, test runner, start command and URL, makes sure the app actually runs, finds login credentials/seed data for UI captures, and drafts .ratchet/standards.md from the codebase if no written standard exists. Spawn

  • ratchet-behavior-reviewer

    Ratchet UI-gate behaviour reviewer. Compares how the real app and the reference respond to the same interaction journey — step-by-step screenshots, accessibility trees, console errors and browser-only test cases — and lists every behavioural difference (wrong result, missing feedback, broken focus,

  • ratchet-checkpoint-planner

    Ratchet checkpoint planner. Splits a large feature, screen migration or refactor into small ordered checkpoints of increasing complexity, each with plain-language "done when" criteria and a decision on which gates apply, written to checkpoints.json. PLAN mode creates the plan; EXTEND mode appends ch

  • ratchet-design-system

    Ratchet design-system agent. Creates or updates DESIGN.md — exact design tokens, component specs with every state, layout, content and accessibility rules, and an implementation map to real files — by extracting from the existing codebase or building from a brief. Follows the ratchet-design-md skill

  • ratchet-fixer

    Ratchet fixer. Takes the adversarial reviewers' open findings for a checkpoint and fixes every one of them — or, if a finding is factually wrong, writes an evidence-backed rebuttal — without touching anything unrelated and without weakening tests. Re-runs the checkpoint tests before returning. Spawn

  • ratchet-implementer

    Ratchet implementer. Writes the production code for exactly ONE checkpoint so its planned tests pass and its UI matches the reference and DESIGN.md — or, when re-spawned with a failing test log or a UI review, fixes exactly those failures. Never edits tests. Spawned by ratchet-orchestrator.

  • ratchet-prototyper

    Ratchet design/prototype agent. Builds a single-file, clickable HTML prototype of the planned feature from DESIGN.md, with every checkpoint's UI state addressable via ?state=<name> and data-testid hooks shared with the real implementation — used to get human/client approval before building and as th

  • ratchet-test-planner

    Ratchet test planner. For ONE checkpoint, plans ID-tagged behavior test cases before any code exists (happy path, edge cases, permissions, failure modes), splitting code-provable cases from browser-only ones, and writes tests/CPn.json plus the checkpoint's tests.files/command. Follows the ratchet-td

  • ratchet-test-runner

    Ratchet test runner. Runs Gate 1 (behavior) for a checkpoint through ratchet.py, and when it fails reads the log and returns a short, precise diagnosis (which case IDs fail, why, where the bug most likely is) so the orchestrator never has to load long test output into its own context. Does not fix a

  • ratchet-test-writer

    Ratchet test writer. Turns a checkpoint's planned test cases (tests/CPn.json) into real, runnable tests tagged with their case IDs, written BEFORE the implementation so every one can run and at least one fails on an assertion. Spawned by ratchet-orchestrator at the start of each checkpoint (sometime

  • ratchet-variant-writer

    Ratchet variant writer. For a high-risk kind of mistake that is about to get an automatic check, writes disguised bad variants of the mistake and look-alike good code, WITHOUT seeing the check, so the check is proven against cases its author didn't choose. Spawned by ratchet-orchestrator at the star

  • ratchet-visual-reviewer

    Ratchet UI-gate visual reviewer. A perfectionist design reviewer that compares screenshots of the real app against the reference (prototype/old screen) in matched states and lists every visual difference — structure, spacing, alignment, size, typography, color, iconography — with severity and on-scr

Hooks (3)

  • Stop

    python3 "${CLAUDE_PLUGIN_ROOT}/skills/ratchet-orchestrator/scripts/ratchet.py" hook stop --host auto

  • UserPromptSubmit

    python3 "${CLAUDE_PLUGIN_ROOT}/skills/ratchet-orchestrator/scripts/ratchet.py" hook prompt --host auto

  • PreToolUse

    python3 "${CLAUDE_PLUGIN_ROOT}/skills/ratchet-orchestrator/scripts/ratchet.py" hook guard --host auto

View source on GitHub

Similar plugins

Matt Pocock's agent skills for real engineering (grilling, spec/ticket flows, TDD, code review, domain modelling and more).

39 skills

mattpocock/skills 323,108

superpowers

claudecodex

Core skills library for Claude Code: TDD, debugging, collaboration patterns, and proven techniques

15 skills · hooks

obra/superpowers 316,836

ecc

claudecodex

Harness-native ECC operator layer - 71 agents, 314 skills, 98 legacy command shims, reusable hooks, rules, selective install profiles, and production-ready workflows for Claude Code, Codex, OpenCode, Cursor, and related agent harnesses

200 skills · 98 commands · 71 agents · 1 MCP server · hooks

affaan-m/ECC 302,088

agent-skills

claudecodex

Production-grade engineering skills covering every phase of software development: spec, plan, build, verify, review, and ship.

25 skills · 4 agents

addyosmani/agent-skills 116,782

Agentic QE (external: maintained in proffesor-for-testing/agentic-qe, pinned to v3.15.17): a quality-engineering fleet with 11 QE agents, 9 skills, 9 commands, the agentic-qe MCP server pinned to the plugin version, and the aqe-mod guard that refuses destructive operations on AQE learning data

9 skills · 9 commands · 11 agents · 1 MCP server

ruvnet/ruflo 81,589

data-agent-kit

claudecodex

This plugin provides a specialized suite of skills for data engineers and database practitioners working on Google Cloud. It acts as an expert assistant, allowing you to use natural language prompts in your preferred coding agent to architect complex data pipelines, transform data with dbt, write Spark and BigQuery SQL notebooks, and orchestrate end-to-end workflows across GCP's data ecosystem.

37 skills

anthropics/claude-plugins-official 47,076

Maintain ratchet-loop?

This page updates itself from your repo. Add the badge to your README, or get it Featured to pin it above the organic results with a followed link to your site.

claude & codex pluginslisted