Vespyr

English | 中文

A platform-agnostic, file-based multi-agent engine. Install 20 specialized AI personas directly into your repository. They plan, design, build, review, test, and ship — with structured memory that persists across sessions and two architectural moats no other framework combines.

📖 Full Documentation →

🏛️ Why Vespyr

| Differentiator | Cost Impact | Quality Impact | |---|---|---| | Socratic methodology depth — every agent declares what it challenges. /grill-me runs an eight-move interrogation frame. | Catches hidden assumptions before code is written — preventing expensive rewrites. | Architectural conflicts and edge cases surfaced by design, not caught in review. | | 3-tier progressive memory — context loads in tiers (core → agent-specific → task-relevant), with proactive pattern pre-fetch. | No context bloat from loading everything. No amnesia from loading nothing. | Agents get relevant past decisions, patterns, and risks without context window flooding. |

Intent & Scope Triage Gate (DNA 4): un-persona'd domain requests are intercepted before execution — ambiguous prompts halt and receive a 2–3 Track Fork card with recommended @agent / /skill handles; trivial tasks pass unimpeded.

Presentation Standards (DNA 6): structured and tabular data uses standard Markdown tables; flows and architecture diagrams use Mermaid; ASCII is strictly reserved for UI wireframes/mockups, CLI previews, and unrenderable graphs.

⚡ Quick Start

# Install globally
npm install vespyr
npx vespyr

# Or install from cloned repo
git clone https://github.com/lalulali/vespyr.git
cd vespyr
npm install
node bin/cli.js --yes --target /path/to/my-project --harness opencode,claude

The interactive CLI wizard configures Vespyr for your IDE harness and bootstraps your project.

# Non-interactive (skip wizard)
npx vespyr --yes --target /path/to/my-project --harness opencode,claude

Then run your first workflow:

/validate-idea   →   /explore-idea   →   /design   →   /develop   →   /launch

New to Vespyr? Follow the step-by-step Getting Started Guide.

👥 The Team (20 Agents)

Core Swarm

| @founder 🧭 | @product-manager 📋 | @product-designer 🎨 | @architect 🏗️ | | @tech-lead 📐 | @developer 💻 | @code-reviewer 🔍 | @qa-engineer 🧪 |

Domain Experts

| @researcher 🔬 | @user-researcher 👥 | @ux-researcher 🎭 | @shifu 🎓 | | @data-analyst 📊 | @security-engineer 🔒 | @performance-engineer ⚡ | @ml-ai-engineer 🤖 | | @ml-ai-ops ⚙️ | @devops-engineer 🚀 | @technical-writer ✍️ |

Memory Layer

@memory-controller 🧠 — script-backed memory service (progressive context loading, validated writes, compaction)

🛠 How It Works

┌──────────────────────────────────────────┐
│               AGENTS (20)                │
│  founder, PM, architect, developer...    │
└──────────────────────────────────────────┘
        │ memory operations
        ▼
┌──────────────────────────────────────────┐
│           MEMORY SERVICE                 │
│  @memory-controller (script-backed)      │
│  3-tier progressive load + pre-fetch     │
└──────────────────────────────────────────┘

Agents act directly. Every persona performs its own reads, writes, and command runs with its own tools; memory operations go through @memory-controller (a script-backed service, not a reasoning-agent delegate).

📋 Pipeline at a Glance

validate → explore → design → plan → develop → launch → iterate → retro

11 phases from idea validation through post-launch iteration. /help-me always tells you what's next. The phase table at .agents/references/phase-table.md is the canonical source.

Full pipeline guide →

🧩 Key Capabilities

• 43 curated skills — from /validate-idea, /design, and /round-table to /shut-up, /motion, /develop, /launch, /customize-skill, and /create-skill (all validated against the agentskills.io spec via CI)
• Consolidated 3-tier memory & machine fencing — atomic state syncing in project-context.md, streamlined 5-file shared memory, and automatic phase-boundary compaction.
• Security & integrity engine — pinned SHA-256 manifest verification (vespyr verify; unsigned interim — accidental-drift detection only, see ADR-002 §2.1.1), supply-chain audit (vespyr audit), closed permission registry, and T3 data trust boundary admission control.
• Agent customization — 2-file TOML merge. Your overrides survive upgrades.
• Multi-harness — first-class installs for OpenCode, Claude Code, Kiro, and GitHub Copilot; the universal .agents/ store is read directly by Aider, Google Antigravity, and Codex.
• 11-phase pipeline — with enforcement. No out-of-order execution.

Complete skills catalog → · Customization →

🔒 Integrity & Security Tooling

# Verify integrity of .agents/ against pinned SHA-256 manifest
npx vespyr verify

# Run supply-chain security and content integrity scanner
npx vespyr audit

# Generate or update .agents/manifest.json checksums
npx vespyr manifest

📚 Documentation

| Level | Resource | |-------|----------| | Start here | Getting Started Guide — first project, invoking agents, basic pipeline | | Comprehensive | Full Guide — installation, configuration, skills, graphs, reference | | Quick overview | QUICK-REFERENCE.md — one-page system cheat sheet | | Agent reference | AGENTS.md — full agent framework & behavioral guidelines |

MIT License. GitHub Issues