| version | 0.1.0 |
|---|
Lightweight spec-driven development for AI coding agents. Spok helps you and your agent agree on what to build, ship one thin slice at a time, and fold the final behavior back into living specs.
Spok is built around four workflow skills you give your AI coding assistant:
/spok-explore -> /spok-propose -> /spok-apply -> /spok-archive
That's the whole surface. Explore is a thinking-only mode for investigating ideas before a proposal. Propose creates a change with a chunked tasks list, apply ships one chunk end-to-end at a time, and archive folds your delta specs into the main specs.
Follow @0xjgv on X for updates.
Requires Bun 1.3.0 or higher.
Install Spok globally:
bun add -g spok@latestThen initialize Spok inside your project:
cd your-project
spok initNow tell your AI coding assistant:
/spok-propose <what-you-want-to-build>
Use /spok-explore <topic> first when you want to think through an idea before proposing work.
spok init configures your AI coding assistants, installs the four workflow skills (spok-explore, spok-propose, spok-apply, spok-archive), and vendors the helper skills they call (spok-flow, spok-create-scoped-chunks, and the rest of the closure).
Note
Not sure if your tool is supported? View the full list. Spok supports 25+ tools and growing.
You: /spok-explore should we add dark mode?
AI: Investigates the existing UI and summarizes options without changing files.
You: /spok-propose add-dark-mode
AI: Created spok/changes/add-dark-mode/
+ proposal.md - why we're doing this, what's changing
+ specs/ - requirements and scenarios
+ design.md - technical approach
+ tasks.md - chunked checklist (3 chunks)
Run /spok-apply to ship the first chunk.
You: /spok-apply
AI: Shipping chunk 1: Add theme context + CSS variables
[runs research -> design -> plan -> implement -> review -> commit]
+ Chunk shipped. 2/3 remaining.
You: /spok-apply
AI: Shipping chunk 2: Wire toggle component to localStorage
+ Chunk shipped. 1/3 remaining.
You: /spok-apply
AI: Shipping chunk 3: Apply theme to remaining surfaces
+ Chunk shipped. 0/3 remaining. Run /spok-archive.
You: /spok-archive
AI: Applied delta specs to spok/specs/ui/spec.md
Archived to spok/changes/archive/2026-05-26-add-dark-mode/
Spok keeps planning artifacts next to your code:
spok/
├── specs/ # Source-of-truth behavior specs
├── changes/
│ └── add-dark-mode/
│ ├── proposal.md # Intent, scope, and approach
│ ├── specs/ # Delta specs for this change
│ ├── design.md # Technical design
│ └── tasks.md # Chunked implementation checklist
└── config.toml # Optional project config
Each change is isolated until you archive it. During archive, Spok applies the delta specs to spok/specs/ and moves the completed change into history.
AI coding assistants are powerful, but they get unpredictable when requirements live only in chat history. Spok adds a lightweight spec layer so the human and agent agree before implementation starts.
- Agree before you build - capture intent, requirements, and design before code changes.
- Stay organized - keep every proposed change in its own folder with specs, design, and tasks.
- Explore before proposing - use
/spok-exploreas a thinking-only mode when the direction is still unclear. - Ship one chunk at a time -
/spok-applyruns a full research -> design -> plan -> implement -> review -> commit loop for one chunk, then stops. - Use your tools - Spok works with 25+ AI coding tools and does not lock you into one IDE or model.
- Getting Started: first steps
- Workflows: combos and patterns
- Commands: slash commands & skills
- CLI: terminal reference
- Supported Tools: tool integrations & install paths
- Concepts: how it all fits
- Multi-Language: multi-language support
- Migration Guide: upgrading from older Spok versions
-> fluid, not rigid
-> iterative, not waterfall
-> easy, not complex
-> built for brownfield, not just greenfield
-> scalable from personal projects to enterprises
vs. Spec Kit (GitHub) - Thorough but heavyweight. Rigid phase gates, lots of Markdown, Python setup. Spok is lighter and lets you iterate freely.
vs. Kiro (AWS) - Powerful but you're locked into their IDE and limited to Claude models. Spok works with the tools you already use.
vs. nothing - AI coding without specs means vague prompts and unpredictable results. Spok brings predictability without the ceremony.
Upgrade the package:
bun add -g spok@latestRefresh agent instructions:
Run this inside each project to regenerate AI guidance and ensure the latest skills are active:
spok updateModel selection: Spok works best with high-reasoning models. We recommend Codex 5.5 and Opus 4.7 for both planning and implementation.
Context hygiene: Spok benefits from a clean context window. Clear your context before starting implementation and maintain good context hygiene throughout your session.
Small fixes - Bug fixes, typo corrections, and minor improvements can be submitted directly as PRs.
Larger changes - For new features, significant refactors, or architectural changes, please submit a Spok change proposal first so we can align on intent and goals before implementation begins.
When writing proposals, keep the Spok philosophy in mind: we serve a wide variety of users across different coding agents, models, and use cases. Changes should work well for everyone.
AI-generated code is welcome - as long as it's been tested and verified. PRs containing AI-generated code should mention the coding agent and model used (e.g., "Generated with Claude Code using claude-opus-4-5-20251101").
- Install dependencies:
bun install - Build:
bun run build - Test:
bun run test - Develop CLI locally:
bun run devorbun run dev:cli - Point global
spokat this checkout:ln -sf "$PWD/bin/spok.js" ~/.bun/bin/spok - Keep the linked CLI current while editing: run
bun run devin one terminal, then usespok ...in another - Conventional commits (one-line):
type(scope): subject
Telemetry
Spok collects anonymous usage stats.
We collect only command names and version to understand usage patterns. No arguments, paths, content, or PII. Automatically disabled in CI.
Opt-out: export SPOK_TELEMETRY=0 or export DO_NOT_TRACK=1
Maintainers & Advisors
See MAINTAINERS.md for the list of core maintainers and advisors who help guide the project.
MIT