REEVES
Stop prompting. Start delegating.
Stop losing context. Start shipping.
curl -fsSL https://raw.githubusercontent.com/Northstar-Technologies/reeves/main/remote-install.sh | bash
BSL 1.1
v3
Claude Code
Stop prompting. Start delegating.
Stop losing context. Start shipping.
curl -fsSL https://raw.githubusercontent.com/Northstar-Technologies/reeves/main/remote-install.sh | bash
Every conversation starts from scratch. Decisions, context, and progress vanish the moment a session ends.
Agents work in isolation. No handoffs, no shared state, no awareness of what other sessions have done.
No token budgets, no model routing. Every job burns Opus tokens whether it needs them or not.
No multi-developer workflows. No way to share agent configurations, context, or session state across a team.
A layered system where every component has a clear role.
22 capabilities that work together as a system, not a grab bag of prompts.
Backend, frontend, security, DevOps, database, and more. Each persona is portable across projects.
/castFile-based event bus for cross-session coordination. Agents share state without sharing context windows.
Automated security scanning dispatched through the Driver system. Clean context, zero contamination.
/scanLead/dev roles, shared INDEX, bus-based coordination. Real teams working the same project with assigned jobs and handoffs.
/onboardRun multiple Claude accounts in parallel as solobot sessions. One human orchestrating a full AI team via the bus.
Token budgets per job. Model routing: Haiku for simple tasks, Sonnet for complex ones, Opus when it matters.
/dispatchJobs, setlists, acceptance criteria, kickoffs, and retrospectives. Real project management, not prompt engineering.
/setlistAdjust verbosity, autonomy, safety, and style. Make Reeves match your workflow, not the other way around.
/tuneGranular control over what agents can do. File access, command execution, and tool usage all scoped per role.
/allowGitHub Issues by default (zero setup), Discord and Slack for real-time. Approve plans, answer questions, and dispatch work from your phone — no terminal needed.
One install command sets up the entire framework. No config files to write, no boilerplate to copy.
/initHandoffs, checkpoints, and bus events. Pick up exactly where you left off, every time.
/goHarvested project context from git, Jira, Slack, and Confluence. Agents start every session informed.
/harvestPass job plans by file path, never by content. Keep conductor context clean and token-efficient.
Free training built in (/train). 9 levels covering fundamentals + lead pipeline. 24 paid specialty tracks.
From idea to shipped code: /pitch → /vision → /scope → /spec → /dispatch → /verify. Or chain them all with /workflow.
MCP server for multi-dev coordination. Live dashboard with agent activity, work tracking, approvals, and human responses — all from a browser.
Two-pass review: devs /scan their own work, leads /verify with holdback scenarios. Tests + acceptance + quality + security.
Define interfaces before implementation. Agents build to contracts, not assumptions.
/specDash Lite for simplicity, Dash Full for heavy orchestration. Same bus events, fully compatible. Pick the mode that fits.
Each role loads only the rules it needs. Conductors, dispatchers, drivers, and sidecars each run with minimal context overhead.
/pitch
CEO / PM
Raw idea → structured pitch
/vision
Architects
Pitch → technical architecture
/scope
Leads
Vision → assignable jobs
/spec
Leads
Job → contract + tests
/dispatch
Leads
Assign to dev or agent
/verify
Leads
Full review + holdbacks
Each command is independent. Use /workflow to chain them all in a guided experience.
Multiple architects can /vision the same pitch independently. Work fans out at every tier. AI agents and human devs receive the same dispatch format.
"Took a partially automated, mostly manual day trading system and in 2 days Reeves output a fully functional automated options trading analysis app."
"Built a complete MSP management system for tracking clients, networks, and infrastructure. I have zero development experience."
"Was burning through my Max x20 plan in two days trying to automate daily tasks with other frameworks. With Reeves I've never hit my usage limit again — and I'm getting more done."
"I just re-factored my entire Fn app into a production ready app with the ability to choose multiple domains and policies within those domains in like an hour. And it works. I followed the Reeves method baby!"
Pick the dispatch mode that fits your setup. Both write identical bus events — fully compatible on the same project.
Reeves spawns agents directly via the Agent tool. No CLI needed, no extra sessions.
Best for: getting started, desktop app users, simple dispatch workflows.
A persistent Dispatcher session handles agent spawning, monitoring, and crash recovery via CLI.
Best for: heavy dispatch (5+ agents), multi-session solo mode, maximum autonomy.
Delivery packages include the job plan, persona assignment, model selection, and token budget. Both modes manage the full lifecycle: spawn, monitor, verify, and park results for your review. You stay in control without micromanaging.
The /harvest command pulls context from your tools and writes it to portable files.
Git-only mode works with zero config. External sources are additive.
decisions.md
Architecture choices, tech stack rationale, and patterns agreed upon by the team.
activity.md
Recent commits, PRs, and changes. Agents know what just happened in the codebase.
conventions.md
Code style, naming patterns, file structure, and team agreements extracted from the project.
sprint.md
Current sprint goals, priorities, and active work items pulled from your project tracker.
Feature-by-feature comparison with other Claude Code frameworks.
| Feature | Reeves | ECC | SuperClaude | Ralph Wiggum |
|---|---|---|---|---|
| Dispatch System | ✔ ★ | ✗ | ✗ | ✗ |
| Event Bus | ✔ ★ | ✗ | ✗ | ✗ |
| Context Engine | ✔ ★ | ✗ | ✗ | ✗ |
| Contract-First Builds | ✔ ★ | ✗ | ✗ | ✗ |
| Multi-Developer | ✔ ★ | ✗ | ✗ | ✗ |
| Remote Control (phone) | ✔ ★ | ✗ | ✗ | ✗ |
| Multi-Account Solo | ✔ ★ | ✗ | ✗ | ✗ |
| Cost Control | ✔ ★ | ✗ | ✗ | ✗ |
| Security Scanning | ✔ ★ | ✗ | ✗ | ✗ |
| Session Persistence | ✔ | ✔ | ✗ | ✗ |
| Job & Setlist Management | ✔ ★ | ✗ | ✗ | ✗ |
| Specialist Agents | ✔ | ✔ | ✔ | ✗ |
| Iterative Loops | ✔ | ✔ | ✗ | ✔ |
| Slash Commands | ✔ | ✔ | ✔ | ✗ |
← scroll to see all columns →
★ = Unique to Reeves
Free training built in — just type /train in any session
Levels 0-6: REEVES Method fundamentals. Levels 7-8: pipeline commands for leads. Progress tracked across sessions.
Deep dives on architecture, advanced dispatch, context engineering, and custom integrations.
11 tracksSecurity scanning, multi-repo workflows, performance optimization, and enterprise patterns.
5 tracksTeam scaling, cost management, org adoption, and measuring AI-assisted development ROI.
8 tracksLearn to apply Reeves to your specific job role.
Advanced training courses starting at $49
Free training built in — type /train in any session
From zero to orchestrated AI agents in under five minutes.
One command sets up the engine repo with all branches.
curl -fsSL https://raw.githubusercontent.com/Northstar-Technologies/reeves/main/remote-install.sh | bash
CLI: cd ~/reeves && claude
Desktop App: Open Project → select ~/reeves
VS Code: Open ~/reeves → Claude panel
New to Reeves? Type /train for the interactive course (9 levels, progress tracked).
Ready to work? Say: "Working on ~/my-project. Fresh start."
Use the pipeline: /pitch → /vision → /scope → /spec → /dispatch → /verify. Or chain it all with /workflow.