Open Specification · Apache 2.0

Stop fighting your AI coding assistant

SCS gives Claude Code the context it needs to work with your system instead of against it. Structured, versioned, git-native. Set up in 15 minutes.

Try the Plugin Star on GitHub

See it in action

Two plugins. Two minutes each.

scs-vibe — Solo Developer

Point it at a PRD or existing project. SCS Vibe scans what's there, asks a few questions, and generates structured context Claude uses every session.

Install scs-vibe →

scs-team — Teams with Docs

Transform your existing PRDs, architecture docs, and compliance checklists into structured context so Claude gets the full picture before writing a line of code.

Install scs-team →

The problem

Claude is good. But it doesn't know your system.

Every session starts fresh. Claude improvises from training data — not your patterns, your constraints, or your team's decisions. The result:

SCS treats context as a first-class artifact — structured, versioned, and git-native, the same way your code is.


Why it works with Claude Code

Built for the tools you're already using

SCS outputs exactly what Claude Code reads: CLAUDE.md and .claude/rules/. No middleware, no new toolchain.

Without SCS With SCS
Context lives in ad-hoc prompts and scattered notes Context is a versioned YAML artifact in your repo
Each session starts from scratch Claude loads structured context from CLAUDE.md every session
Team members give Claude different instructions Context is shared, reviewable, and consistent across the team
No audit trail for what Claude was told Git history shows every context change and who made it

For regulated industries

The standard and the system it governs never meet

Governance frameworks like CHAI, HIPAA, and SOC2 are written for human readers. AI systems run on structured data and explicit constraints. Nothing connects the two — until now. SCS includes a dedicated standards bundle type so that governance requirements travel directly from the standards body to the AI agent, intact and versioned.

CHAI T&E Framework

15 requirements across fairness, accuracy, scope, and transparency — transposed into a machine-readable reference bundle.

Read the article →

Prior authorization AI

What changes when your coding agent has the compliance framework in its context before writing a line of criteria-matching logic.

Read the article →

HIPAA · NIST · SOC2

Standards bundles are composable. A prior auth agent can load CHAI, HIPAA, NIST cybersecurity controls, and internal policy as a single unified governance stack.

See the spec →

Reference implementation

The CHAI standards bundle is open source in the SCS repository — a working example of what a governance standard looks like when an AI agent can actually use it.

View on GitHub →

If you represent a standards organization interested in publishing your framework in SCS format, get in touch.


From the blog

On context engineering

A structured context stack with a healthcare shield representing governed AI criteria for prior authorization workflows

The Governance Gap in Prior Authorization AI

March 4, 2026

PA AI has largely solved data connectivity. The governed criteria problem — versioned, auditable rules that travel with each agent — remains unsolved, and regulators are about to make that gap very visible.

Layered isometric diagram showing Claude Code's context hierarchy — Enterprise Managed Policy at the base, Global in the middle, Project with CLAUDE.md and .claude/rules/ at the top

Claude Code Has a Context Architecture. You're Probably Only Using Part of It.

March 3, 2026

Claude Code ships with a layered context hierarchy — enterprise, global, project, and rules. Most developers only use part of it. SCS is the principled way to fill in the rest.

A layered structured context stack with an authorized context shield defending against prompt injection attacks

Beyond Guardrails: A Structural Defense for AI Agent Security

March 3, 2026

Filters and trained refusals address symptoms. Structured context addresses the gap — an authoritative reference point the agent can compare incoming instructions against.

All articles →


Documentation

Reference and guides


If SCS is solving a real problem for you, star the repo — it helps other developers find it.