Back

archē

Archē (pronounced R-K) is an ancient Greek concept meaning: the first principle, origin, or fundamental foundation from which everything arises.

Archē is a vibecoding framework or "meta-context" that gives Claude everything it needs to build your app autonomously and efficiently.

context/ folder structure

Everything Claude knows about your project lives in context/. Generated during setup and kept up to date automatically. You never need to re-explain your project

context
  • design
  • developer
  • features
  • ops
  • project
  • technical
  • README.md
  • SETUP.md
CommandWhat it does
/setupStart here. Runs the setup wizard to understand your idea and generate all project context files. Run once on a new project.
/continuePick up where you left off. Claude reads the project state, figures out what was in progress, and resumes - no explanation needed. Use this any time you start a new session or feel lost.
/build [feature]Build a feature end-to-end - spec, code, review, tests, browser testing, and docs. Example: /build user login
/fix [bug]Fix a bug. Describe what's wrong in plain English and Claude investigates and fixes it. Example: /fix the signup form isn't sending emails
/test [feature]Generate a browser test guide for any feature. Leave blank to test the most recently built feature.
/review [file]Run a code quality review. Leave blank to review recent changes.
/statusSee the current state of the project - active tasks, feature progress, recent work, and what's next.
/tasksView and manage the task list. See active, blocked, and completed tasks.
/tasks nextFind out exactly what to work on next.
/tasks add "..."Add a new task manually.
/tasks done T5Mark a task as complete.
/syncSync all context files to reflect the current state of the codebase. Run if the docs feel out of date.
/deep [problem]Invoke the Opus AI agent for complex bugs or architecture decisions. Claude will warn you about the higher cost before proceeding.

Claude uses a set of specialist agents behind the scenes - each focused on a specific job. They're dispatched automatically; you never need to invoke them directly.

AgentWhat it doesCost
scope-checkerChecks whether a request fits the agreed project plan before any work startsVery low
feature-plannerTurns a feature idea into a detailed spec with user stories, edge cases, and a task breakdownLow
code-reviewerReviews code for quality, security, and correctness before it's committedMedium
test-writerWrites automated tests for completed featuresMedium
uat-guideGenerates the plain-English browser test checklist for you to followMedium
context-updaterKeeps all context files and the task list up to date after work completesVery low
next-actionReads the task list and roadmap to determine what to work on nextVery low
deep-solverDeep investigation for complex bugs or architecture decisions. Uses the Opus model - requires your confirmation due to higher costHigh

When you run /build, here's exactly what happens:

1
Scope check
Claude verifies the feature is within the agreed project plan before touching any code.
2
Feature spec
If no spec exists, Claude creates one - a breakdown of what the feature does, who uses it, the happy path, edge cases, and a task list. You review it before anything is built.
3
Build
Claude works through the task list one task at a time, marking each one complete as it goes. You see real progress, not just a finished result at the end.
4
Code review
An automatic review checks for quality issues, security problems, and anything that doesn't match your project's conventions.
5
Automated tests
Claude writes and runs tests. If any fail, it fixes them before continuing.
6
Browser testing ← the human step
Claude generates a friendly checklist of things to try in your actual browser. You follow the steps and report back. If anything looks wrong, describe it - Claude fixes it. This is the only step that requires you to do anything.
7
Context sync
All documentation is updated: task list, feature status, architecture notes, any decisions made during the build.

Start a new project with Claude and include the context folder in the root. Run /start to begin.