spec-first

Spec-first agentic development is specification-driven development for AI coding agents using reviewed specs, isolated worktrees, and gated commits.

Spec-first agentic development is a specification-driven workflow for AI coding agents. Four commands write the spec, check it against the code, build it in an isolated worktree, and deliver it through commit checks. A fresh reviewer checks each spec against the code before the build starts, and a gate runs verify stages before commits land. State lives in context.yaml, making the system resumable after restarts. It supports ticket providers like GitHub and Jira to pull work into the pipeline. The method works with Claude Code, Codex, Cursor, Copilot, and through a git commit-msg hook.

spec-first at a glance

Best for: Software engineer
Runs on: Web, CLI, API, GitHub, Jira, Claude Code, Cursor, Copilot
Pricing: Free
Released: October 2026
Category: AI coding tool
Website: spec-first.dev

Key features

Write the spec from a description
Review specs against existing code
Build features in isolated worktrees
Gate commits with verify stages
Deliver pull requests in parallel
Log decisions with reasoning next to code
Bootstrap projects from existing code

Use it for

Pulling tickets from GitHub and Jira into agent pipelines
Building features in isolated worktrees in parallel
Running automated checks before commits land
Logging architectural decisions alongside code

Alternatives to spec-first

cc-sdd — Explicit boundaries and independent review let humans and agents work in parallel safely.
Spec Kit — Provides structured processes and templates for AI coding agents.
SpecDeck — It gives human teams and coding agents a single shared source of truth to build from.
Agentic Coding Biolerplate — Provides shared instructions and review workflows for coding assistants.
Agent Skills — Encodes senior-engineer workflows and quality gates into 24 skills.
Specops — It stops AI from coding without requirements by enforcing a multi-phase spec workflow.
Agentic Bug Hunter — It drives the reconnaissance, triage and reporting loop an agent would run.
AGENTS.md — It separates agent-specific instructions from human-focused README files.
Reviewer Agent — It uses a separate AI model to review code and prevent premature agent stopping.
GitHub Agentic Workflows — Automates repository maintenance safely using AI agents in GitHub Actions.
Specweave — It coordinates multiple AI tools and agents with shared specs, memory, and an append-only ledger.
CodeGraph — It eliminates file crawling, reducing AI agent tool calls by 88% and cutting costs.

All spec-first alternatives