The open standard & high-performance execution engine for AI-driven software engineering. Write a machine-readable spec — let your agents plan, build, verify, and remember.

Topologically sorts your dependency graph in < 2ms, compiles precise context bundles, and orchestrates agents through the full lifecycle.

Autonomous agent skill registration, category-based discovery, skill invocation, rating systems, and real-time cost tracking.

Decoupled event mesh topic routing and message broadcasting across distributed autonomous swarm nodes.

Tasks aren't 'done' until their verify scripts exit 0. Failures are marked [!] Blocked, halting the engine.

Expose your project's live architecture to Claude Desktop, Cursor, and any MCP client with tools like alp_get_graph.

49 JSON schemas registered with 1:1 TypeScript & Python SDK parity and 1013+ passing automated tests.

Git standardized version control. Docker standardized environments. OpenAPI standardized APIs. ALP standardizes how AI builds software.

Today every AI coding assistant (Devin, Claude Code, Cursor, OpenHands) relies on unstructured prompts and brittle context-scraping. They forget decisions, overwrite each other's work, and lose track of dependencies. ALP replaces scattered README.md, PRD.md, AGENTS.md, and TASKS.md files with one deterministic protocol stored natively in your repository (.alp/).

ALP turns your repository into a deterministic, machine-readable project specification that any AI agent can read, understand, and act on.

flowchart LR ALP[.alp/ Files] --> Parser[ALP Parser] Parser --> Graph[Dependency Graph] Graph --> Engine[Execution Engine] Engine --> Agent[AI Agent] Agent --> Task[Task Execution] Task --> Verify[Quality Gates] Verify --> State[Project State]

Write — Define your project in .alp/ files using the ALP protocol

Parse — The parser builds a dependency graph and validates against 49 JSON schemas

Execute — The engine topologically sorts tasks and compiles context bundles in < 2ms

Verify — Quality gates ensure tasks aren't marked done until tests pass

Remember — Cross-session memory eliminates redundant context scraping

Quick Example ​

Create a .alp file in your repository:

!alp-version: 80.0.0

@project id: my-project name: My Project version: 1.0.0 state: active

@feature id: feat-auth name: Authentication description: OAuth2 + JWT authentication flow

@task id: task-login-ui name: Build login UI status: [ ] depends_on:

• task-setup-db verify:
• npm run test:login
• npm run lint:login

@agent id: agent-frontend name: Frontend Specialist capabilities: [react, typescript, tailwind]

Then run:

Validate your workspace

alp validate

Execute the next available task

alp run

Verify quality gates

alp verify task-login-ui

How it works (detailed) ​

1. The Execution & Verification Cycle ​

ALP parses your workspace into a Directed Acyclic Graph (DAG). Agents only receive the exact context they need, exactly when they need it.

graph TD subgraph Repository [Your Repository .alp/] D[Decision: Use PostgreSQL] --> T1[Task: Setup DB Schema] T1 --> T2[Task: Build REST API] R[Rule: No Raw SQL] --> T2 C[Contract: @contract c-api] --> T2 end

subgraph ExecutionEngine [Execution Engine - alp run] T2 -->|Context Bundle - 1.8ms| Agent[Claude / Cursor Agent] end

subgraph QualityGates [Quality Gates - alp verify] Agent -->|npm test| V{Tests Pass?} V -->|Exit 0| X[Mark x Done] V -->|Non-Zero| B[Mark ! Blocked] end

2. Autonomous Swarm Mesh & Skill Marketplace (v80.0.0) ​

mermaidflowchart LR subgraph SwarmNodes [Autonomous Swarm Nodes] A1[Agent Alpha\nCoder] <--> EM((Event Mesh\nPub/Sub)) A2[Agent Beta\nReviewer] <--> EM A3[Agent Gamma\nTester] <--> EM end

subgraph Marketplace [Swarm Marketplace] EM <--> SWM[Skill Registry\n@swarm_marketplace] SWM -->|Discover & Invoke| Cost[Cost & Metering Engine] end

The ALP Ecosystem ​

@autonomous-lifecycle-protocol-alp/cli

Terminal CLI interface: run, marketplace, event-mesh, policy, vault, verify.

Parses .alp files and computes Kahn topological sorts over the dependency graph in sub-2ms.

Real-time MCP integration for Claude Desktop, Cursor, and any compliant client.

Language Server with IntelliSense, go-to-definition, and rich hover metadata.

Official TypeScript and Python SDKs with complete 1:1 implementation parity.

Official Go, Rust, and Java SDKs with core parsing, graph, and workspace APIs.

Autonomous skill registry, provider discovery, invocation, and cost tracking (v38.0.0).

The unified ALP desktop IDE (Mac/Windows/Linux). Faster, more secure, and error-free vs. fragmented multi-IDE setups. Native @autonomous-lifecycle-protocol-alp/parser, Monaco editor, agent manager, IDE Intelligence with AI suggestions/predictions/review, and Autonomous Orchestration with self-healing DAGs.

npm install -g @autonomous-lifecycle-protocol-alp/cli

Step 2: Initialize your workspace ​

bashalp init

This creates a .alp/ directory with a starter project.alp file.