Skip to content

Corpusvision/agent-framework

Repository files navigation

Agent Framework CLI

npm version License: MIT Node.js TypeScript

Specification-driven agents for every AI editor — one command, all platforms.

agent-framework-cli installs a governed engineering workflow into your project and emits the right agent files for whichever AI editor you use: GitHub Copilot, Cursor, Claude Code, Windsurf, or any Open Plugins-compatible tool.


One command. All platforms.

npm install -g agent-framework-cli
acli setup

acli setup asks two questions (project type + memory backend), installs all dependencies, and scaffolds your project. Open your AI editor and start:

/acli.run Build a user authentication system with OAuth support

Platform Support

acli init asks which editors you target. Agents are emitted to all selected destinations automatically.

Platform Output location Format
GitHub Copilot .github/agents/{name}.agent.md VS Code agent frontmatter
Cursor .cursor/rules/{name}.mdc YAML frontmatter + alwaysApply
Claude Code AGENTS.md Aggregated markdown sections
Windsurf .windsurf/rules/{name}.md YAML frontmatter + trigger
Open Plugins .agents/plugins/agent-framework/agents/{name}.agent.md Open Plugins spec

Every acli install / acli remove writes to all configured platforms and keeps the Open Plugins manifest (.plugin/plugin.json) in sync.


The Power Stack

Tool What it does
spec-kit Spec-driven development — keeps every feature grounded in structured requirements
Fleet Multi-agent orchestration — architect, security, dev, and QA run in the right order
Superpowers Bridge Connects your editor to TDD enforcement, brainstorming, and structured code review
Beads (optional) Semantic memory — architecture decisions and task state survive every session
Open Plugins Canonical extension format — share and install custom plugins across any conformant tool

Why bother?

AI coding assistants write code fast but without discipline. Left unchecked:

  • Features get built from vague prompts instead of structured specs
  • Architecture drifts because every session starts fresh
  • Tests get skipped, security gets overlooked, reviews get bypassed

Agent Framework wraps your AI editor in a governed engineering workflow — spec before code, review before merge, memory that persists.


The Lifecycle

Run /acli.run <feature> — the orchestrator drives everything. You only step in at three approval gates.

# Phase What happens
1 Specify Constitution + feature spec authored
2 Clarify Ambiguities resolved before any planning
3 Plan Architecture decisions + implementation plan
4 Checklist Security, a11y, and performance gates generated
5 Tasks Prioritised, dependency-ordered task breakdown
🚦 Gate 1 You approve the plan before any code is written
6 Analyze Cross-artifact consistency validated
🚦 Gate 2 You confirm everything is aligned
7 Review Cross-model plan review
8 Implement Dev + QA loop — up to 5 iterations before escalation
9 Verify Full test suite + acceptance criteria checked
10 Finish Branch cleanup, merge readiness confirmed
🚦 Gate 3 You give final approval before merge

5 Agents, 1 Team

Agent Job
orchestrator Routes work, enforces gates, runs tasks in parallel
architect Writes specs, plans, ADRs — tech-agnostic and rigorous
security OWASP checklists, threat modeling, vulnerability review
development TDD implementation, plan-conformant, hands off cleanly
qa Code review, test coverage, acceptance validation

Slash Commands

Everyday:

Command What it does
/acli.run <feature> Full lifecycle start to finish
/acli.onboard Analyse and bootstrap an existing codebase
/acli.implement Jump straight to implementation

Spec & Planning:

Command What it does
/acli.constitution Define your project principles & architecture boundaries
/acli.specify Author a structured feature spec with acceptance criteria
/acli.plan Generate implementation plan + architecture decision records
/acli.tasks Break the plan into prioritised, dependency-ordered tasks
/acli.checklist Generate security, a11y, and performance quality gates

Review & Ship:

Command What it does
/acli.critique Spec-aligned code review with severity ratings
/acli.debug Structured root-cause analysis
/acli.finish Branch cleanup and merge readiness check

Agency Roster

Install any of 140+ community-built agents from the Agency roster — organized across engineering, design, marketing, product, and more.

# Browse all available agents by division
acli agency list

# Narrow to a specific division
acli agency list --division engineering

# Search by keyword
acli agency search "data engineer"
acli agency search "frontend"

# Install by slug or keyword
acli agency install engineering-frontend-developer
acli agency install frontend-developer          # matches by keyword
acli agency install engineering/frontend-developer  # division-scoped

# Force overwrite an existing install
acli agency install engineering-frontend-developer --force

Agents are installed to all platforms in .agent-framework.json and the index is cached locally for 24 hours. Use --refresh to fetch the latest roster.

Platform Output path
GitHub Copilot .github/agents/{slug}.agent.md
Cursor .cursor/rules/{slug}.mdc
Claude Code .claude/agents/{slug}.md
Windsurf .windsurf/rules/{slug}.md
Open Plugins .agents/plugins/agency-agents/agents/{slug}.agent.md

Extension Ecosystem

Extend the framework with additional agents, skills, and rules via the acli extensions command.

# See what is installed and what is available
acli extensions list --available

# Install a known extension
acli extensions add brownfield          # speckit brownfield onboarding
acli extensions add fleet               # multi-agent fleet orchestration

# Install any custom open-plugins package
acli extensions add my-plugin --from https://example.com/my-plugin.zip --ecosystem open-plugins

# Build and share your own plugin
acli extensions create my-plugin       # scaffold structure + starter agent + skill
acli extensions pack my-plugin         # zip for distribution

Supported extension ecosystems

Ecosystem Install mechanism Discovery
speckit specify extension add CLI spec-kit extension registry
open-plugins Download + extract zip .agents/plugins/*/plugin.json manifest

Workspace Layout

.github/
  agents/         ← GitHub Copilot agent definitions
  skills/         ← reusable skills
  prompts/        ← slash commands
.cursor/
  rules/          ← Cursor rule files (.mdc)
.windsurf/
  rules/          ← Windsurf rule files (.md)
.agents/
  plugins/        ← Open Plugins packages (manifest + agents/skills/rules)
AGENTS.md         ← Claude Code agent registry (auto-generated)
.specify/
  memory/         ← constitution, architecture, quality standards
  specs/          ← specs, plans, checklists, tasks
.beads/           ← persistent memory (if beads-based memory selected)
.agent-framework.json  ← framework config (platforms, agents, dependencies)

CLI Reference

Command Description
acli setup [--check] Install dependencies + scaffold project (interactive)
acli init [--platforms …] Initialise framework, select target platforms
acli install <agent> Install a framework agent to all configured platforms
acli remove <agent> Remove an agent from all configured platforms
acli list agents|skills List available or installed agents/skills
acli update [agent] Re-emit agents to all configured platforms
acli config View or modify framework configuration
acli agency install <name> Install a community agent from the Agency roster
acli agency list [--division] Browse 140+ community agents by division
acli agency search <query> Search community agents by keyword
acli extensions list List installed extensions
acli extensions add <name> Install an extension
acli extensions remove <name> Remove an extension
acli extensions create <name> Scaffold a new custom plugin
acli extensions pack <name> Package a plugin as a distributable zip

Requirements

  • Node.js >= 18
  • Python >= 3.8 (for spec-kit)
  • One or more supported AI editors: GitHub Copilot (VS Code), Cursor, Claude Code, or Windsurf

Contributing

See CONTRIBUTING.md. MIT licensed.

Supercharge your agentic development — stop fiddling with agent config and start shipping.

agent-framework-cli sets up everything you need to follow a perfect SDLC lifecycle with GitHub Copilot, in one command. Spec → Plan → Implement → Review → Ship, with quality gates and persistent memory baked in.


One command. Full setup.

npm install -g agent-framework-cli
acli setup

That's it. acli setup asks you two quick questions, installs all dependencies, and scaffolds your project. Open Copilot Chat and go:

/acli.run Build a user authentication system with OAuth support

The Power Stack

Agent Framework wires together four best-in-class tools so you don't have to:

Tool What it does for you
spec-kit Spec-driven development engine — keeps every feature grounded in structured requirements, not vibes
Fleet Multi-agent orchestration — runs architect, security, dev, and QA agents in the right order with the right context
Superpowers Bridge Connects Copilot to brainstorming, TDD enforcement, and structured code review workflows
Beads (optional) Semantic memory across sessions — your architecture decisions, ADRs, and task state survive every conversation

Choose Git-based memory (zero extra tooling, context in plain files) or Beads-based memory (advanced semantic search + chunking). acli setup asks you at install time.


Why bother?

AI coding assistants write code fast but without discipline. Left unchecked:

  • Features get built from vague prompts instead of structured specs
  • Architecture drifts because every session starts fresh
  • Tests get skipped, security gets overlooked, reviews get bypassed
  • Nobody knows why a decision was made two weeks ago

Agent Framework wraps Copilot in a governed engineering workflow — spec before code, review before merge, memory that persists.


The Lifecycle

Run /acli.run <feature> — the orchestrator drives everything. You only step in at three approval gates.

# Phase What happens
1 Specify Constitution + feature spec authored
2 Clarify Ambiguities resolved before a line of planning
3 Plan Architecture decisions + implementation plan
4 Checklist Security, a11y, and performance gates generated
5 Tasks Prioritised, dependency-ordered task breakdown
🚦 Gate 1 You approve the plan before any code is written
6 Analyze Cross-artifact consistency validated (spec ↔ plan ↔ tasks)
🚦 Gate 2 You confirm everything is aligned
7 Review Cross-model plan review
8 Implement Dev + QA loop — up to 5 iterations before escalation
9 Verify Full test suite + acceptance criteria checked
10 Finish Branch cleanup, merge readiness confirmed
🚦 Gate 3 You give final approval before merge

5 Agents, 1 Team

Agent Job
orchestrator Routes work, enforces gates, runs tasks in parallel
architect Writes specs, plans, ADRs — tech-agnostic and rigorous
security OWASP checklists, threat modeling, vulnerability review
development TDD implementation, plan-conformant, hands off cleanly
qa Code review, test coverage, acceptance validation

Slash Commands

Everyday:

Command What it does
/acli.run <feature> Full lifecycle start to finish
/acli.onboard Analyse and bootstrap an existing codebase
/acli.implement Jump straight to implementation

Spec & Planning:

Command What it does
/acli.constitution Define your project principles & architecture boundaries
/acli.specify Author a structured feature spec with acceptance criteria
/acli.plan Generate implementation plan + architecture decision records
/acli.tasks Break the plan into prioritised, dependency-ordered tasks
/acli.checklist Generate security, a11y, and performance quality gates

Review & Ship:

Command What it does
/acli.critique Spec-aligned code review with severity ratings
/acli.debug Structured root-cause analysis
/acli.finish Branch cleanup and merge readiness check

Brownfield? No problem.

/acli.onboard auto-scans your existing codebase, generates a tailored constitution + reference architecture, and gets you into the lifecycle without starting from scratch.


Workspace Layout

.github/
  agents/    ← agent definitions
  skills/    ← reusable skills
  prompts/   ← slash commands
.specify/
  memory/    ← constitution, architecture, quality standards
  specs/     ← specs, plans, checklists, tasks
.beads/      ← persistent memory (if beads-based memory selected)

CLI Reference

Command Description
acli setup [--check] Install dependencies + scaffold project (interactive)
acli install <agent> Install a specific agent
acli remove <agent> Remove an installed agent
acli list agents|skills List available agents or skills
acli update [agent] Update agents, prompts, and skills
acli config View or modify framework configuration

Requirements

  • Node.js >= 18
  • VS Code + GitHub Copilot Chat
  • Python >= 3.8 (for spec-kit)

Contributing

See CONTRIBUTING.md. MIT licensed.

About

Turn GitHub Copilot into a coordinated engineering team — specification-driven development lifecycle with quality gates, persistent memory, and human-in-the-loop governance

Topics

Resources

License

Contributing

Stars

Watchers

Forks

Packages

 
 
 

Contributors