Claude Code builds before it thinks.
Guild makes it spec first.

Spec-first workflows and specialized agent roles for Claude Code. Markdown files, tracked by git, zero infrastructure.

Guild council demo
View on GitHub
scroll

Three commands. Full pipeline.

From zero to spec-first development in minutes. Guild installs the workflows, Claude Code runs them.

1 guild init

Interactive onboarding. Installs spec-first workflows and agent roles into your project.

2 /guild-specialize

Explores your codebase. Enriches CLAUDE.md with real context, conventions, and patterns.

3 /council

Multiple perspectives evaluate your idea independently. A design doc captures the decision before code is written.

Spec first, then build

Every feature starts with structured deliberation. Code comes after the design doc. Claude Code executes each phase natively.

/build-feature "Add JWT auth"
Evaluate
evaluate
Design
spec + plan
Implement
implement
Reviewer
review
QA
test
Bugfix
fix

Six roles. Clear boundaries.

Instead of one AI doing everything, six specialized agents each handle what they're best at. Claude Code reads the roles and assumes them natively.

Advisor

Evaluates ideas and gives strategic direction. First gate before any work begins.

Tech Lead

Breaks features into tasks with acceptance criteria. Defines technical approach, architecture, and implementation direction.

Developer

Implements features following project conventions, writes tests, makes atomic commits.

Code Reviewer

Reviews code quality, patterns, and tech debt. Guards the standards of the guild.

QA

Testing, edge cases, and regression. Validates that the implementation meets acceptance criteria.

Bugfix

Diagnosis and bug resolution. Isolates root causes and applies targeted fixes.

Structure without infrastructure

Spec before code

Every feature starts with a design doc. Requirements, trade-offs, and decisions are captured before a single line of code is written.

Structured deliberation

Multiple agents analyze your idea independently, in parallel. Diverse perspectives converge into a clear, well-reasoned decision.

Nothing lost between sessions

SESSION.md captures where you stopped. Claude Code's memory captures what you learned. Guild's session skills write to both — you resume with full context of what you know and what you were doing.

Zero infrastructure

No servers, no databases, no API keys. Just markdown files and Claude Code. Works in any project, any language, any stack.

Measurable skill quality

guild eval validates that your skills route correctly, follow expected structure, and don't regress. Built-in benchmarking with accuracy, precision, recall, and regression detection.

Get started in 30 seconds

Install as a Claude Code plugin or via npm.

/plugin marketplace add Guild-Agents/guild
/plugin install guild
or
$ npx guild-agents init