v2.0.0
GitHub
Open Source · MIT License

BTRS Documentation

24 specialist AI agents. 24 skills. Discipline enforced.

AI agent orchestration with discipline enforcement (TDD, verification, debugging), session continuity, and the Superpowers merge. /btrs picks up where you left off and enforces the Iron Laws on every task.

Get Started →

One Command

Type /btrs and describe what you need. The router handles the rest.

🤖

24 Specialists

Architects, engineers, security, ops, and business agents — each with deep domain knowledge.

🛠

Convention-Aware

Auto-detects your project patterns. Agents follow YOUR style, not generic templates.

Self-Verifying

Every agent checks its own work with evidence before reporting done.

📚

Obsidian Vault

All decisions and documentation auto-written to a linked knowledge base.

🔀

Any AI Tool

Designed for Claude Code. Agent definitions work with other AI tools (Cursor, Copilot, Windsurf) but the full workflow requires Claude Code.

Quick Start

Get BTRS running in under 60 seconds.

Claude Code
# Install BTRS
git clone https://github.com/btrs-ai/btrs.git ~/.claude/btrs && ~/.claude/btrs/install.sh

# Then in any project:
/btrs build me a user dashboard
💡
Tip

First time in a project? BTRS auto-scans your codebase. No setup needed.

Installation

BTRS supports all major AI development tools. Pick yours below.

Install via script:

bash
git clone https://github.com/btrs-ai/btrs.git ~/.claude/btrs && ~/.claude/btrs/install.sh

Symlinks all skills and agents into ~/.claude/. Run install.sh again to update.

ℹ️
Note

After install, /btrs is available globally in every project you open with Claude Code.

BTRS is designed for Claude Code. For Cursor, you can reference the agent definitions at agents/btrs-*/AGENT.md and the universal instructions in AGENTS.md. The full skill workflow (subagent dispatch, TDD enforcement, session continuity) requires Claude Code.

BTRS is designed for Claude Code. For GitHub Copilot, you can reference the agent definitions at agents/btrs-*/AGENT.md and the universal instructions in AGENTS.md. The full skill workflow (subagent dispatch, TDD enforcement, session continuity) requires Claude Code.

BTRS is designed for Claude Code. For Windsurf, you can reference the agent definitions at agents/btrs-*/AGENT.md and the universal instructions in AGENTS.md. The full skill workflow (subagent dispatch, TDD enforcement, session continuity) requires Claude Code.

BTRS is designed for Claude Code. For OpenAI Codex, you can reference the agent definitions at agents/btrs-*/AGENT.md and the universal instructions in AGENTS.md. The full skill workflow (subagent dispatch, TDD enforcement, session continuity) requires Claude Code.

Upload AGENTS.md as project context in Claude AI. Load individual agent files (agents/btrs-*/AGENT.md) as needed for specialized tasks.

ℹ️
Note

The web interface does not support slash commands. Paste agent instructions directly into your conversation context.

First Run

When you use /btrs in a project for the first time, the system will:

Scan your project

BTRS reads your file structure, dependencies, and existing patterns to build a conventions profile.

Generate conventions

A btrs/conventions/ folder is created with your project's detected patterns — naming, structure, style, and anti-patterns.

Initialize the vault

The btrs/ Obsidian-compatible vault is set up for decision logs, ADRs, and documentation.

Ready to work

Every subsequent /btrs call in this project uses the conventions automatically. No repeated setup.

How It Works

You describe what you need

Type /btrs followed by a natural language request. No special syntax required.

BTRS classifies the request

The router analyzes your request to determine which domain(s) it belongs to — architecture, frontend, security, ops, etc.

Specialist agents are dispatched

The appropriate agents are loaded with your project's conventions injected as mandatory context. Complex tasks may involve multiple agents in sequence.

Agents verify their own work

Each agent runs a 5-point self-verification check before reporting completion. If anything fails, the agent self-corrects.

Documentation is auto-updated

Decisions, changes, and rationale are written to the btrs/ Obsidian vault automatically.

The /btrs Command

The /btrs command is the universal entry point. Here are examples of different request types:

examples
# Building
/btrs build a notification system with email and push

# Fixing
/btrs the login page crashes on mobile Safari

# Planning
/btrs plan the Q2 product roadmap

# Security
/btrs audit the API for vulnerabilities

# DevOps
/btrs set up CI/CD with GitHub Actions

# Research
/btrs compare Postgres vs CockroachDB for our use case

# Documentation
/btrs document the payments API with examples
💡
Tip

You can be as specific or as vague as you want. BTRS will ask clarifying questions if needed.

Follow-Up Conversations

After your initial /btrs command, you do not need to prefix every message. Just keep talking naturally.

conversation
# Start with /btrs
/btrs build a user settings page

# Then just follow up normally
Add a dark mode toggle to that page
Actually, make it a dropdown with 3 options: light, dark, system
Now add tests for the theme switching logic

The active agent retains context from the conversation. Use /btrs again when you want to switch to a completely different task.

Agents — Management

1 Agent
💼

Boss Agent

Central coordinator. Breaks down requests, dispatches agents, tracks progress.

/btrs build a full-stack app/btrs plan the migration to microservices

Agents — Technical

4 Agents
🏗

Architect

System design, architecture decisions, ADRs, technology selection.

/btrs design the auth system/btrs propose a caching strategy
🧪

QA & Test Engineering

Unit, integration, e2e, performance, and security testing.

/btrs write tests for the payment flow/btrs set up e2e testing with Playwright
📝

Documentation

Technical writing, API docs, guides, changelogs.

/btrs document the REST API/btrs write the onboarding guide
🔬

Research

Technology evaluation, competitive analysis, innovation scouting.

/btrs compare React vs Vue for our needs/btrs research real-time sync solutions

Agents — Engineering

6 Agents
🔌

API Engineer

Backend APIs, REST, GraphQL, microservices, server logic.

/btrs build a REST API for orders/btrs add GraphQL subscriptions
🌐

Web Engineer

React, Vue, frontend applications, SPAs, SSR.

/btrs build a dashboard with React/btrs fix the responsive layout
📱

Mobile Engineer

React Native, Flutter, iOS, Android development.

/btrs build the onboarding screens/btrs fix push notifications on iOS
🖥

Desktop Engineer

Electron, Tauri, native desktop applications.

/btrs build a system tray app/btrs add auto-update functionality
🎨

UI Engineer

Component libraries, design systems, Storybook, visual polish.

/btrs create a button component system/btrs build the design tokens
🗃

Database Engineer

Schema design, optimization, migrations, query tuning.

/btrs design the schema for multi-tenancy/btrs optimize the slow queries

Agents — Security

2 Agents
🔐

Code Security

SAST/DAST, vulnerability scanning, secure coding practices, OWASP compliance.

/btrs audit the auth code for vulns/btrs add input sanitization
🛡

Security Ops

Infrastructure security, compliance (GDPR, SOC2, HIPAA), access controls.

/btrs review our AWS IAM policies/btrs prepare for SOC2 audit

Agents — Operations

5 Agents

Cloud Ops

AWS, Azure, GCP, infrastructure as code, networking.

/btrs set up the staging environment/btrs migrate from Heroku to AWS
🚀

CI/CD Ops

Deployment pipelines, GitHub Actions, automated releases.

/btrs set up CI/CD with GitHub Actions/btrs add preview deploys for PRs
📦

Container Ops

Docker, Kubernetes, orchestration, container security.

/btrs dockerize the application/btrs set up Kubernetes manifests
📈

Monitoring Ops

Prometheus, Grafana, alerting, observability, log aggregation.

/btrs set up monitoring and alerts/btrs add distributed tracing
🛠

DevOps

Combined DevOps workflows, deployment orchestration, platform engineering.

/btrs set up our DevOps workflow/btrs orchestrate the deployment pipeline

Agents — Business

6 Agents
📊

Product

Product strategy, roadmap, requirements, user stories.

/btrs write user stories for onboarding/btrs plan the Q2 roadmap
📣

Marketing

Campaigns, growth strategies, SEO, content marketing.

/btrs write the launch email sequence/btrs optimize our landing page SEO
💰

Sales

Sales strategy, pipeline optimization, revenue analysis.

/btrs build the sales pitch deck outline/btrs analyze the conversion funnel
💲

Accounting

Financial management, bookkeeping, revenue tracking.

/btrs set up the billing system logic/btrs model the subscription pricing tiers
🤝

Customer Success

Retention strategies, support workflows, NPS tracking.

/btrs design the customer onboarding flow/btrs build a churn prediction model
📐

Data Analyst

Business intelligence, analytics, dashboards, data modeling.

/btrs build a KPI dashboard/btrs analyze user engagement metrics

Skills — Core

3 Skills
SkillDescriptionWhen to Use
/btrsUniversal router. Classifies your request, resumes active work, and dispatches the right agent(s).Starting any new task or resuming a session. This is your default entry point.
/btrs-initInitializes BTRS in the current project. Scans conventions, creates vault.First time using BTRS in a project (auto-runs on first /btrs).
/btrs-dispatchRoutes work to specialist agents with context injection and progress tracking.Used internally by the pipeline. Power users can invoke directly.

Skills — Planning

4 Skills
SkillDescriptionWhen to Use
/btrs-brainstormExplores intent, requirements, and design before implementation. Mandatory before any creative work.Automatically invoked before building features. Also available directly.
/btrs-planCreates a detailed implementation plan with task breakdown and agent assignments.Before starting complex, multi-step work.
/btrs-proposeGenerates a proposal with multiple solution approaches (minimal, conventional, scalable).When you need to evaluate options before committing to an approach.
/btrs-specWrites a technical specification with requirements, constraints, and acceptance criteria.Before building a new feature or system.

Skills — Building

4 Skills
SkillDescriptionWhen to Use
/btrs-executeExecutes implementation plans task-by-task with TDD enforcement and review checkpoints.When you have a plan and are ready to build. Replaces manual agent dispatching.
/btrs-implementImplements a plan or spec. Dispatches engineering agents with conventions enforced.When you have a plan and are ready to build.
/btrs-worktreeCreates isolated git worktrees for feature work. Keeps main branch clean.Before starting any feature branch work. Auto-invoked by the pipeline.
/btrs-deployHandles deployment configuration, CI/CD setup, and release management.When code is ready to ship.

Skills — Discipline

3 Skills
SkillDescriptionWhen to Use
/btrs-tddTest-driven development enforcement. Red → Green → Refactor cycle on every implementation.Automatically invoked during execution. Can also be used standalone for any coding task.
/btrs-debugSystematic 4-phase debugging: observe, hypothesize, test, conclude. No guessing allowed.Any bug, test failure, or unexpected behavior.
/btrs-sanity-checkRuns all tests, linters, and checks. Produces evidence of pass/fail status.After implementation, before finishing. Auto-invoked by the pipeline.

Skills — Workflow

4 Skills
SkillDescriptionWhen to Use
/btrs-finishCompletes a development branch: merge, PR, or cleanup with structured options.When implementation is done and all tests pass.
/btrs-request-reviewRequests code review before merging. Produces structured review request.After completing a feature, before merge.
/btrs-receive-reviewProcesses code review feedback with technical rigor. No blind agreement.When receiving review comments. Verifies suggestions before implementing.
/btrs-handoffCreates a context handoff document for continuing work in a new session.End of session. Passing work to another developer.

Skills — Quality

4 Skills
SkillDescriptionWhen to Use
/btrs-reviewCode review with convention compliance, best practices, and security checks.Before merging code. After implementation.
/btrs-auditDeep security and compliance audit of code and infrastructure.Before releases. Periodic security checks.
/btrs-verifyVerifies convention compliance across the project. Catches drift.Periodically, or after large changes.
/btrs-healthProject health check — dependency freshness, tech debt, test coverage.Regular maintenance. Sprint planning.

Skills — Knowledge

4 Skills
SkillDescriptionWhen to Use
/btrs-researchDeep research on technologies, patterns, or business topics with citations.Evaluating new tools. Making technology decisions.
/btrs-analyzeData analysis, metrics review, and insights generation.Understanding user behavior. Measuring impact.
/btrs-docGenerates or updates documentation (API docs, guides, READMEs).After building features. Onboarding new team members.
/btrs-tech-debtTech debt lifecycle — scan, triage, fix, report.Sprint planning. Reducing accumulated debt. Maintenance cycles.

Workflow Pipeline

Every task in BTRS v2.0.0 follows a structured pipeline. No shortcuts, no skipped steps.

/btrs
Classify
Brainstorm
Plan
Worktree
Execute
TDD per task
Sanity-Check
Finish

Classify

BTRS analyzes your request and determines the type: feature, bugfix, question, research, or multi-domain task.

Brainstorm

Explores your intent, requirements, and design constraints before touching any code. Uses /btrs-brainstorm.

Plan

Creates a detailed implementation plan with task breakdown, agent assignments, and acceptance criteria.

Worktree

Isolates work in a git worktree so your main branch stays clean. Uses /btrs-worktree.

Execute with TDD

Each task is implemented test-first: write failing test, write minimal code to pass, refactor. Uses /btrs-execute and /btrs-tdd.

Sanity-Check

Runs all tests, linters, and verification checks. Uses /btrs-sanity-check.

Finish

Merges work, cleans up worktree, updates documentation. Uses /btrs-finish.

💡
Tip

For simple questions or quick lookups, BTRS skips the full pipeline and answers directly from conventions.

Discipline — The Three Iron Laws

BTRS v2.0.0 enforces three non-negotiable laws on every task. These cannot be skipped or overridden.

🧪

1. TDD or No Code

Every implementation starts with a failing test. No test, no code. Enforced by /btrs-tdd. Red → Green → Refactor on every task.

2. Evidence Before Assertions

Never claim "it works" without proof. Show test output, build logs, or command results. Verification is mandatory, not optional.

🔎

3. Systematic Debugging

No guessing. Bugs are debugged through a 4-phase process: observe, hypothesize, test, conclude. Uses /btrs-debug.

⚠️
Non-Negotiable

These laws are enforced automatically. Agents cannot skip TDD, cannot claim success without evidence, and cannot guess at bug fixes. This is what makes BTRS output reliable.

Session Continuity

Start a new session and /btrs picks up where you left off. No manual handoffs needed.

example
# New session — BTRS detects active work
/btrs

# Output:
Active work detected in btrs/work/:
  Feature: user-notifications (3/5 tasks complete)
  Worktree: feature/user-notifications
  Next task: Build push notification service
Resume? [Y/n]

BTRS tracks active work in the btrs/work/ directory. Plans, progress, and evidence are all persisted so sessions can resume seamlessly across conversations.

Convention System

Conventions are how BTRS prevents AI drift — the tendency for AI tools to generate code that ignores your project's existing patterns.

How Conventions Work

Auto-detected from your project

BTRS scans file structure, naming patterns, imports, component styles, and configuration to build a conventions profile specific to YOUR project.

Injected into agent context

Conventions are loaded as mandatory context for every agent dispatch. Agents cannot ignore them — they are part of the prompt.

Component registry prevents duplicates

A registry of existing components, utilities, and services means agents reuse what exists instead of recreating it.

Anti-pattern lists prevent mistakes

Known bad patterns from your codebase are explicitly flagged so agents avoid them.

/btrs-verify catches violations

Run /btrs-verify to scan the entire project for convention violations and get automated fix suggestions.

⚠️
Important

Conventions are project-specific. Each project gets its own conventions profile. Moving between projects automatically switches context.

Self-Verification

Every agent performs a 5-point verification check before reporting work as complete. This is not optional — it is built into every agent's instructions.

The 5-Point Check

  • Convention compliance — Does the output follow all detected project conventions?
  • Functional correctness — Does the code compile, run, and produce correct results?
  • Integration safety — Does it break any existing functionality?
  • Security baseline — Are there any obvious security issues (injection, exposure, etc.)?
  • Evidence provided — Can the agent show proof of verification (test output, build success, etc.)?
💡
Tip

If any check fails, the agent fixes the issue and re-verifies. You only see the final, verified result.

Multi-Solution Exploration

When you use /btrs-propose, the system generates three distinct approaches to solve your problem, ranked by complexity and scalability.

The 3-Approach Pattern

🟢

Approach A: Minimal

Fewest changes. Solves the immediate problem. Low risk, fast to ship. Best when time is critical.

🔵

Approach B: Conventional

Industry-standard solution. Balanced effort and quality. Good documentation and community support.

🟣

Approach C: Scalable

Future-proof architecture. Higher upfront cost but handles 10x growth. Best for core systems.

Each approach includes estimated effort, tradeoffs, and a recommendation. You pick the one that fits your situation.

Obsidian Vault

BTRS v2.0.0 uses a btrs/ directory (no dot prefix) with a three-tier structure: knowledge, work, and evidence. It is plain Markdown, compatible with Obsidian for visual graph exploration.

Three-Tier Directory Structure

btrs/
  config.json
  knowledge/            # Persistent project knowledge
    conventions/
    decisions/
    code-map/
    tech-debt/
  work/                 # Active session state
    specs/
    plans/
    todos/
    changelog/
    status.md
  evidence/             # Verification proof
    reviews/
    verification/
    debug/
    sessions/
  .obsidian/
ℹ️
Note

You do not need Obsidian installed. The vault is plain Markdown files. Obsidian just gives you a nice visual graph view of how decisions and documentation are interconnected.

Cross-AI Compatibility

BTRS is designed for Claude Code. The agent definitions in agents/btrs-*/AGENT.md and the universal instructions in AGENTS.md can be referenced by other AI tools, but the full skill workflow requires Claude Code.

.cursorrules
.windsurfrules
AGENTS.md
Universal Source
copilot-instructions.md
CLAUDE.md

Each adapter file translates the universal agent instructions into the format expected by that platform. When you update AGENTS.md, the adapters stay in sync.

PlatformAdapter FileAuto-loads?
Claude CodeCLAUDE.md + ~/.claude/ symlinksYes
Cursor.cursorrulesYes
GitHub Copilot.github/copilot-instructions.mdYes
Windsurf.windsurfrulesYes
OpenAI CodexAGENTS.md (native)Yes
Claude AI (Web)Manual uploadNo

FAQ

Do I need to type /btrs before every message?
No. Type it once to start a task, then just keep talking naturally. The active agent retains context from the conversation. Use /btrs again when you want to switch to a completely different task.
Does it work with my existing project?
Yes. /btrs auto-scans your project on first use. No configuration file to write, no setup wizard. It detects your conventions from the codebase itself.
What if I don't use Claude Code?
AGENTS.md works with any AI tool. Cursor, Copilot, Windsurf, and Codex all have adapter files that auto-load. For tools without native support, upload the agent files as context manually.
Do I need Obsidian?
No. The btrs/ vault is plain Markdown files that work with any text editor. Obsidian just provides a nice visual graph view and linked note browsing. It is entirely optional.
Can I use individual agents without the /btrs router?
Yes. Load any agent's AGENT.md file as context in your AI tool and interact with it directly. The /btrs router is a convenience layer, not a requirement.
How do I update?
If installed as a plugin, it auto-updates. For manual installs, run ~/.claude/btrs/install.sh again — it pulls the latest and re-links everything.
How do I uninstall?
Run ~/.claude/btrs/uninstall.sh. It removes all symlinks from ~/.claude/ cleanly. Your projects are not affected.
Does it cost anything?
No. BTRS is free and open source under the MIT License. You just need access to an AI tool (Claude Code, Cursor, etc.) which may have its own subscription costs.
Will agents overwrite my code style?
No. Conventions are auto-detected from YOUR project. Agents follow your existing patterns — naming conventions, file structure, component styles, and import patterns. They adapt to you, not the other way around.
What happens if an agent makes a mistake?
Every agent self-verifies with evidence before reporting done. The 5-point check covers convention compliance, functional correctness, integration safety, security baseline, and evidence. If any check fails, the agent fixes the issue and re-verifies automatically.