Everything Claude Code

How the repo is split

Three folders, three jobs.

If a file tells an agent who it is and which tools it may use, it is an agent. If it describes a workflow the agent should follow, it is a skill. If it is a slash command you type to start that work, it is a prompt. Public HTTP APIs are a different library on this site.

• Agents

  agents/

  A delegated specialist with a name, a description, and a tool list. Use one when the task has a boundary, such as review or a failing build.

• Skills

  skills/

  Domain knowledge and a method. The agent or a command loads it. A skill is not the same file as the agent that follows it.

• Prompts

  commands/

  Slash commands such as /plan and /tdd. They are the words you send to start a workflow. They are not the public API catalogue.

• Not APIs

  mcp-configs/

  MCP server entries let an agent call GitHub, a host, or a database. That is an agent connection. For a list of public HTTP APIs, use the Public APIs page.

Agents

Who each specialist is for.

• Before code

  planner

  You asked for a feature, a refactor, or an architectural change and want a plan before edits.

• Before code

  architect

  You are deciding structure, scale, or a technical tradeoff, not implementing the feature yet.

• While coding

  tdd-guide

  You want tests written first, then the smallest code that passes, with coverage kept high.

• After a change

  code-reviewer

  You just wrote or modified code and want a quality, security, and maintainability pass.

• After a change

  security-reviewer

  The change handles user input, authentication, an API, or sensitive data.

• When red

  build-error-resolver

  The build or TypeScript check failed. You want the smallest diff that makes it green, not a redesign.

• User flows

  e2e-runner

  You need Playwright journeys, artifacts, and a check that a critical path still works.

• Cleanup

  refactor-cleaner

  You want unused code and duplicates found and removed, with tests still passing.

• Docs

  doc-updater

  Codemaps, READMEs, or guides no longer match the code.

Skills

The workflow the agent should follow.

• Daily code

  coding-standards

  Shared TypeScript, JavaScript, React, and Node habits for the whole codebase.

• Server

  backend-patterns

  API shape, database access, and caching for Node, Express, and Next.js routes.

• Browser

  frontend-patterns

  React and Next.js UI, state, and performance patterns.

• Analytics data

  clickhouse-io

  ClickHouse query and data-engineering patterns. Skip it if you do not run ClickHouse.

• Security

  security-review

  A checklist for auth, user input, secrets, new endpoints, and payments.

• Tests

  tdd-workflow

  The method behind the TDD agent: failing test, minimal code, then refactor.

• Trust

  verification-loop

  A check on the current session before you treat the result as done.

• Trust

  eval-harness

  A formal eval loop for sessions, for when a single manual check is not enough.

• Long sessions

  strategic-compact

  Suggests compacting context at a real task boundary instead of at a random limit.

• Long sessions

  continuous-learning

  Pulls a repeated pattern out of a session and saves it as a skill for later.

• Template

  project-guidelines-example

  An example of a project-specific skill. Copy the shape. Do not treat the sample as your project’s rules.

Prompts

Slash commands you type.

These files live in commands/. They are prompts that start a workflow. They do not replace the agent or the skill they call.

• /plan

  Plan, then wait

  Restate the requirement, name the risks, and write the steps. Do not edit code until you confirm.

• /tdd

  Tests first

  Scaffold the interface, write the failing tests, then the smallest implementation.

• /code-review

  Review the diff

  A security and quality pass on uncommitted changes.

• /build-fix

  Fix the build

  Work through TypeScript and build errors without a wide rewrite.

• /e2e

  Browser journeys

  Generate and run Playwright tests and keep the artifacts.

• /refactor-clean

  Remove dead code

  Find unused code and remove it only with tests to check the result.

• /verify

  Check the tree

  Run the verification loop against the current codebase.

• /checkpoint

  Save a point

  Create or check a checkpoint so a long task has a known state.

• /eval

  Eval workflow

  Manage eval-driven checks when you are scoring sessions, not just reading them.

• /learn

  Save a pattern

  Look at this session and extract a pattern worth keeping as a skill.

• /orchestrate

  Sequence agents

  Run a sequential agent workflow when one specialist is not the whole task.

• /setup-pm

  Package manager

  Set npm, pnpm, yarn, or bun for the hooks. Also /test-coverage, /update-codemaps, and /update-docs for coverage and doc sync.