AI-SDLC
The decision engine for autonomous software development
| Rev. | Date | Note |
|---|---|---|
| v1alpha1 | early |
- Symbol
- Ai
- No.
- 01
- Group
- Fire
- State
- Live
spec v1alpha1 (early) - Updated
An open-source framework that governs how AI coding agents turn human decisions into shipped software. Humans answer open decisions up front, answers become a spec, and agents implement under quality gates. Spec version v1alpha1.
How adoption works
Decisions
Your people answer the open decisions up front. The Decision Engine routes each open question to the right person, with a recommendation and a counter-argument.
Spec
The answers become a spec: declarative resources validated by JSON Schema. A Definition-of-Ready gate holds the work until the spec is complete.
Gated agents
An orchestrator dispatches developer subagents into isolated worktrees. They implement and open pull requests under graduated quality gates and tiered autonomy.
Review
A reviewer on a different harness from the implementer checks the work, and the review is sealed with a signed Proof of Execution. Your team reads the specs; the line-by-line review is the reviewer’s job.
Getting it running
- Claude Code plugin
/ai-sdlc setup- Any MCP client
npx -y @ai-sdlc/mcp-advisor- SDKs
@ai-sdlc/sdk · ai-sdlc-framework · Go
Walk one task through it
Pick a sample task and answer the three questions an agent would otherwise guess at. Your answers write the spec, and the spec decides what the agent builds and what it is allowed to do on its own.
Decisions
The Decision Engine routes each open question to a person, with a recommendation and a counter-argument.Definition of Ready: 0 of 3 decisions answered. Dispatch is held until all 3 open ones are answered.
Spec
Waiting on the decisions at 10.Gated agents
Waiting on the spec at 20.
What it governs
Everything is a declarative resource, so a pipeline is reviewed like code. Six resources carry the model:
Pipeline- The path work takes, from an answered decision to a reviewed pull request.
Decision- An open question, routed to a person with a recommendation and a counter-argument.
AgentRole- What an agent does in the pipeline, such as implementer or reviewer.
QualityGate- A graduated check the work has to pass before it moves on.
AutonomyPolicy- How much an agent may do on its own, set in tiers.
AdapterBinding- Which coding agent fills a role: Claude Code, Copilot, Cursor, Codex or Aider.
Around the resources: a usage ledger with model routing, an operator console for pending decisions, the dependency graph and usage, and compliance mapping to the EU AI Act, NIST AI RMF and ISO 42001.
Status and maturity
- State
- Live and open source.
- Spec
- v1alpha1. Early: expect the resource shapes to change before v1.
- Activity
- 3,077 commits on main. 448 stars and 81 forks on GitHub (as of 2026-11-19).
- Adoption figures
- None published. The statistics on ai-sdlc.io come from third-party studies of the problem, and none of them measure adoption of AI-SDLC.
- Docs
- Onboarding, spec primer, API reference, tutorials, FAQ and a self-check at ai-sdlc.io.
The builders
- Dominique LegaultChief Technology Officer, Arcana Advisory
The engineer behind AI-SDLC.
Adopting it on your team?
Bring your pipeline and the places your agents keep getting stuck. Dominique will tell you plainly whether it fits, and if it does, you adopt it with the people who built it.
Talk to the team about adopting AI-SDLC