description
spec-kit
Spec-driven development engine — keeps every feature grounded in structured requirements, acceptance criteria, and architecture decisions. Spec before a line of code is written.
Read the Docs →Open Source · MIT LicenseNode.js ≥ 18 · VS Code + Copilot
agent-framework-cli
A governed engineering workflow for GitHub Copilot. One CLI installs everything — spec-driven development, multi-agent execution, human approval gates, and persistent memory. Spec before code. Review before merge. Ship with confidence.
terminal — setup
# Install globally
$ npm install -g agent-framework-cli
+ agent-framework-cli@2.0.0
# Run setup (2 questions, done in ~30 sec)
$ acli setup
? Memory backend: (git-based / beads) › git-based
? Enable Beads for semantic memory? › No
✓ spec-kit installed
✓ Fleet agents configured (5 agents)
✓ Superpowers bridge ready
✓ Project scaffolded
# Open Copilot Chat and run:
/acli.run "Add rate limiting to the payments API"
The Problem
AI writes code fast. Without discipline, that's a liability.
Left unchecked, AI coding assistants accumulate technical debt faster than your team can review it. Agent Framework adds the guardrails enterprises need without slowing developers down.
closeVague prompts instead of structured specs
closeArchitecture drifts — every session starts fresh
closeTests skipped, security overlooked, reviews bypassed
closeNo audit trail of why decisions were made
The Power Stack
Four best-in-class tools, wired together.
group_work
Fleet
Multi-agent orchestration — runs architect, security, dev, and QA agents in the right order with the right context. Parallel execution with enforced human approval gates.
electric_bolt
Superpowers Bridge
Connects Copilot Chat to brainstorming, TDD enforcement, and structured code review. Turns a general-purpose assistant into a governed engineering tool.
memory
Beads (optional)
Semantic memory across sessions — architecture decisions, ADRs, and task state persist across every conversation. Choose git-based (zero deps) or beads-based at install time.
The Lifecycle
Run /acli.run — the orchestrator drives everything.
You only step in at three approval gates. Every other step is automated — spec, plan, security review, implementation, tests, and merge readiness.
lock_open
Gate 1
You approve the plan before any code is written
lock_open
Gate 2
You confirm cross-artifact alignment
lock_open
Gate 3
You give final approval before merge
description
1Specify
Constitution + feature spec authored with acceptance criteria
chat_bubble
2Clarify
Ambiguities resolved before any planning begins
architecture
3Plan
Architecture decisions + implementation plan generated
checklist
4Checklist
Security, a11y, and performance quality gates generated
task_alt
5Tasks
Prioritised, dependency-ordered task breakdown
lock_open
G1Gate 1
You approve the plan before any code is written
analytics
6Analyze
Cross-artifact consistency validated (spec ↔ plan ↔ tasks)
lock_open
G2Gate 2
You confirm everything is aligned
rate_review
7Review
Cross-model plan review before a single line of code
code
8Implement
Dev + QA loop — up to 5 iterations before escalation
verified
9Verify
Full test suite + acceptance criteria checked
merge_type
10Finish
Branch cleanup, merge readiness confirmed
lock_open
G3Gate 3
You give final approval before merge
5 Agents, 1 Team
Specialized agents for every phase.
manage_accounts
orchestrator
Routes work, enforces gates, runs tasks in parallel
architecture
architect
Writes specs, plans, ADRs — tech-agnostic and rigorous
security
security
OWASP checklists, threat modeling, vulnerability review
code
development
TDD implementation, plan-conformant, hands off cleanly
fact_check
qa
Code review, test coverage, acceptance validation
Slash Commands
Run directly from Copilot Chat.
boltEveryday
/acli.run <feature>
Full lifecycle start to finish — the one you'll use most
/acli.onboard
Analyse and bootstrap any existing codebase
/acli.implement
Jump straight to implementation (skips spec/plan)
descriptionSpec & Planning
/acli.constitution
Define 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
rate_reviewReview & Ship
/acli.critique
Spec-aligned code review with severity ratings
/acli.debug
Structured root-cause analysis
/acli.finish
Branch cleanup and merge readiness check
CLI Reference
Terminal commands.
acli setup [--check]
Install dependencies + scaffold project
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
Workspace Layout
What gets scaffolded.
project root
.github/
agents/ ← agent definitions
skills/ ← reusable skills
prompts/ ← slash commands
.specify/
memory/ ← constitution, architecture
specs/ ← specs, plans, tasks
.beads/ ← persistent memory
(beads-based only)check_circleNode.js ≥ 18
check_circleVS Code + GitHub Copilot Chat
check_circlePython ≥ 3.8 (for spec-kit)
Ready to ship better software, faster?
Open source and free. Enterprise version with Foundry integration, dedicated support, and on-prem deployment available.