Workspine skill

A workflow for coding agents: a plan you approve, evidence-based verification and a handoff of state between sessions.

  • 58 ★ on the repository
  • Last commit
  • MIT
Skill

Extended access

Why this level

  • The install creates service folders in the repository, and the skills direct the agent to edit code and run checks
All reasons and checks

patricksys/workspine

Install

In your terminal, with SkillFoxx CLI

npx skillfoxx add skills/workspine

Detects the agents on your machine, shows the access level and pins the version.

Other ways to install

Access and checks

How we review

Extended accessThe tool runs code or makes network calls and only reads external data.

  • The install creates service folders in the repository, and the skills direct the agent to edit code and run checks

This is third-party code. Review the repository files before installing.

What it does

Workspine brings agent work into one order: plan first, then execution, then verification, with everything recorded in the repository. The `npx -y workspine setup` command adds a `.work/` folder with durable project state and `.agents/skills/work-*` skills to the current repository. Agent-specific adapters are optional. Work goes through skills. `work-plan` builds a checked plan and asks questions about what matters, after which you give approval in chat and the agent records it against the plan. `work-execute` implements the approved plan and `work-verify` checks the result by evidence: if the plan makes a UI claim, it can require real browser evidence. There are also `work-quick` for an understood small change, `work-new-project` for a start, and `work-pause`, `work-resume` and `work-progress` for handing state to a new session.

Who it is for. Developers running a project with several agents and sessions who want decisions and checks kept with the code.

Good fit when

  • An agent guesses for you too quickly and a behavior decision must be approved up front
  • Tests passed but the rendered interface is still wrong and a browser check is needed
  • Work stopped mid-task and a new session must continue from where it left off
  • Different agents change the project and shared state must live in the repository

Not a fit when

  • A task is self-contained with obvious decisions and checks, and the agent’s normal workflow is enough
  • Service folders `.work/` and `.agents/` cannot be added to the repository
  • Fully autonomous work with no questions or approvals is needed

Example request

Use work-plan: a docs sidebar should reveal the selected page without breaking manual collapse. Show the questions and wait for my approval.

Limitations

Node.js 22 or newer is required. In headless environments the install runs as `npx -y workspine setup --yes`, giving explicit permission for bounded writes. Codex CLI is the best proven runtime: one end-to-end run is recorded only for it, Claude Code and OpenCode receive the same generated files without an end-to-end run, and Cursor, GitHub Copilot and Gemini CLI are claimed as qualified support. The `main` branch may contain unreleased changes, and the version is checked with `npx -y workspine --version`. The installer writes files into the repository.

How to disable. Delete the `.work/` and `.agents/skills/work-*` folders from the repository, and the agent adapters if installed. A global install is removed separately per the docs.

FAQ

How do I continue after stopping?

Before stopping call `work-pause` and check that decisions, unfinished work and the next action were saved. In a fresh session run `work-resume` or `work-progress`.

When should I use work-quick?

When the change is already understood and needs less ceremony. For a project that still needs shaping there is `work-new-project`.

What is kept in `.work/`?

The decision, an implementation summary and verification evidence. A fresh session recovers decisions and unfinished work from these records.

How do I approve a plan?

Say explicitly in chat that you agree. The agent records the approval against the plan and confirms it, with no manual ID or record edit.

Which agents are supported?

Codex CLI best. Claude Code and OpenCode receive the same files, and Cursor, GitHub Copilot and Gemini CLI work with caveats.

Discussion

No questions or reviews yet.

Skills

Builder Skills

Skills for agents

A set of small skills from Builder.io for agent work habits: visual plans and diff recaps, docs-first checks, cross-agent review and clean status updates

SkillExtended access4.7KRepository stars

Nine JS Mastery skills take a change through phases: from idea and spec to verification, tests, release texts and a docs sync

SkillExtended access1.6KRepository stars

Plans, decisions and verification results live in Git: another agent or a teammate resumes the task with ~go

SkillExtended access369Repository stars

Nine short skills: tests, review, ADR plans, slides, a wiki, image generation, worktrees and plan implementation.

SkillExtended access59Repository stars

A plugin for Claude Code and Codex with xdo, xdel and xflow modes: a main agent plans, a coder builds, an evaluator checks.

PluginExtended access54Repository stars
Russian stack

Orchestrator skill: from an idea the agent writes requirements and a spec, splits work into tasks, builds the project, checks it

SkillBroad access611Repository stars

README in short

The process needs no separate service: everything lives in the repository and commands are called from the agent chat. Verification relies on evidence, not the agent’s claim.

README badge

Report a problem