DecisionLayer

Turn bulky specs into the few decisions a developer must review before AI starts coding.

DecisionLayer is an open source pre-implementation control layer for AI-driven development. It reads the spec artifacts a team already has, extracts the decisions and ambiguities that actually matter, predicts likely agent assumptions, and emits agent guardrails before implementation begins.

Why this exists

Spec-driven development is producing more markdown, not more clarity. Developers increasingly face:

• spec fatigue from long, repetitive documents
• implicit technical decisions buried in prose
• AI agents that fill gaps with risky assumptions
• reviews that happen after code already exists

DecisionLayer exists to preserve developer control at the exact moment it matters most: right before an AI agent turns ambiguous prose into code.

Product goal

Help developers stay in control of AI-driven implementation by turning large specs into a small set of reviewable decisions, risks, and constraints before an agent starts coding.

Core loop

1. Read the active spec bundle.
2. Extract the top implementation decisions, risks, and ambiguities.
3. Show the developer a short decision preflight.
4. Produce an agent-guardrails.yaml artifact.
5. Let the agent implement under explicit constraints.

Principles

• Simplicity wins.
• Decision-first, not summary-first.
• Nearby beats separate dashboards.
• Output must shape agent behavior.
• Action beats completeness.

Repository layout

.
├── .github/                  # community health, issue templates, workflows
├── docs/                     # product doctrine, architecture, roadmap, ADRs
├── examples/                 # sample spec bundles and expected outputs
├── packages/
│   ├── cli/                  # local CLI surface for repo-first workflows
│   └── core/                 # parser, extraction, scoring, and output logic
├── schemas/                  # public schemas for DecisionLayer artifacts
├── CONTRIBUTING.md
├── CODE_OF_CONDUCT.md
├── GOVERNANCE.md
├── LICENSE
├── ROADMAP.md
└── SECURITY.md

What this repo includes today

This repository starts intentionally small:

• a clear product doctrine
• a contributor-friendly repo structure
• a public schema for agent-guardrails
• a sample spec bundle and expected output
• placeholder package boundaries for core logic and CLI integration

What we are not building first

• a generic markdown beautifier
• a project management suite
• a giant spec dashboard
• support for every framework before proving the core loop
• a fully autonomous coding agent

Near-term build target

The first real product milestone is a local decision preflight that reads a spec bundle and emits:

• concise brief
• top 3 to 5 implementation decisions
• top unresolved ambiguities
• top risky agent assumptions
• readiness verdict
• agent-guardrails.yaml

Community

Community contribution is welcome. See CONTRIBUTING.md, CONTRIBUTING_PRODUCT.md, GOVERNANCE.md, and ROADMAP.md.

If you want to understand the product principles behind the project, start with docs/product/principles.md, docs/product/first-principles.md, and docs/product/decision-framework.md.

License

Apache-2.0. See LICENSE.