Skip to content

speclabs/devspec

Repository files navigation

devspec

devspec is a spec-driven development framework for teams using GitHub Copilot and other AI coding agents.

It stores planning, implementation, review, and recovery state in Git-tracked files instead of relying on chat history. Use it when you want agents to follow the same workflow for project context, engineering standards, work-item intake, implementation, review, and session recovery.

Quick Start

Install with the devspec CLI from the target repository. The recommended one-off path is uvx, which avoids a permanent global install and works well on restricted developer machines.

uvx devspec init --target . --profile all --repo-state existing
uvx devspec doctor --target . --profile all

Use --repo-state new for a new repository. After validation, commit the copied framework files, run the foundation flow, then start the first work item.

For the full setup manual, see docs/how-to/README.md.

Setup Options

Situation Start here
Simplest one-off setup uv and uvx setup
Persistent macOS/Linux package install Homebrew setup
Approved Windows package install WinGet setup
CLI setup paths are blocked Manual copy setup

Install profiles let you choose which AI coding agent files to copy. Most multi-agent teams can start with --profile all; single-tool teams can use a smaller profile such as copilot, codex, cursor, claude, gemini, or antigravity.

Core Workflow

devspec has two workflow layers.

Layer Purpose Full guide
Foundation Capture stable project context, architecture, stack, structure, standards, and rules. New repository, existing repository
Work items Move one feature, bug, security issue, or accepted change request from intake to review. Work-item lifecycle

For a new project:

New repository foundation flow

/devspec.projectcontext
/devspec.techstack
/devspec.codebase-structure
/devspec.coding-standards
/devspec.rules

For an existing project:

Existing repository foundation flow

/devspec.extract
/devspec.projectcontext
/devspec.techstack
/devspec.codebase-structure
/devspec.coding-standards
/devspec.rules

Then run the work-item flow:

Work-item lifecycle flow

/devspec.story
/devspec.finalize
/devspec.tasks
/devspec.implement
/devspec.review

Use /devspec.clarify only when a work item records a blocking question. Use /devspec.diagram when a diagram would clarify architecture, workflow, state, sequence, or domain behavior.

AI Tool Support

GitHub Copilot prompt and agent files are the reference implementation. Other adapters are thin wrappers around the same command registry and Git-tracked artifacts.

Tool Setup notes
GitHub Copilot Native /devspec.* prompt commands through .github/prompts/ and .github/agents/.
Claude Code Project skills expose /devspec-* command-style invocations.
OpenAI Codex AGENTS.md provides always-on repository instructions and Codex workflow guidance.
Cursor Project rules guide Cursor Agent and Inline Edit.
Gemini CLI GEMINI.md provides context; optional commands expose /devspec:* shortcuts.
Google Antigravity Workspace rules and skills expose /devspec-* command-style invocations.

Canonical command names remain /devspec.*. Some adapters expose host-native shortcuts such as /devspec:story or /devspec-story; see Command Invocation by Agent.

More Documentation

Need Start here
Install or troubleshoot setup docs/how-to/setup/README.md
Run devspec workflows docs/how-to/README.md
Use command examples docs/how-to/README.md#command-examples
Set up AI coding agents docs/how-to/README.md#ai-coding-agent-setup
Work across multiple repositories docs/how-to/README.md#multi-repo-work
Upgrade devspec files docs/how-to/README.md#upgrades
Validate adapter behavior devspec/adapters/validation-flows.md
Review adapter support details devspec/adapters/README.md
Check platform compatibility devspec/adapters/compatibility-matrix.md
Plan enterprise rollout devspec/adapters/enterprise-governance.md

License

This repository is released under the Apache License 2.0.

About

Enterprise-ready spec-driven development framework that helps software teams and technical leaders standardize AI-assisted delivery across architecture, planning, implementation, review, governance, and recovery.

Topics

Resources

License

Stars

1 star

Watchers

0 watching

Forks

Contributors