Sopify skill
Plans, decisions and verification results live in Git: another agent or a teammate resumes the task with ~go
- 369 ★ on the repository
- Last commit
- Apache-2.0
Extended access
Why this level
- The installer runs through `curl | bash` and appends a managed block to global instruction files such as `~/.claude/CLAUDE.md`
- On the `~go` command the agent runs the plan and changes project files, and writes verification results into the repository
Install
- Works in
Manual install
curl -fsSL https://github.com/evidentloop/sopify/releases/latest/download/install.sh | bash -s -- --target codex:en-USThere are seven install targets, each with an `en-US` and a `zh-CN` variant. Read the script before running it.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The installer runs through `curl | bash` and appends a managed block to global instruction files such as `~/.claude/CLAUDE.md`
- On the `~go` command the agent runs the plan and changes project files, and writes verification results into the repository
This is third-party code. Review the repository files before installing.
What it does
Sopify is a set of rules and skills that installs into an AI coding tool. When a task needs a plan, the agent saves the plan, the decisions made and the verification evidence as ordinary files in `.sopify/` and commits them with the code. Questions, code reading and small fixes are handled directly, without a plan. There are three commands. `~go plan <request>` creates a plan: the agent first asks about anything unclear and writes `.sopify/plan/<date>_<topic>/plan.md` only after your answers, changing no code. `~go` runs the current plan or resumes it one task at a time: implement, verify, review, mark done. `~go finalize` closes the plan and moves it to `.sopify/history/`. The state lives in the repository, so interrupted work can be picked up by another agent or a person: a fresh Qoder session needed only `~go` to continue a plan written in Claude Code. For several agents, the docs describe the roles of plan owner, executor and reviewer.
Who it is for. Developers and teams who run long tasks in Codex, Claude Code or Cursor and lose context when the session or the tool changes.
Good fit when
- A long task was interrupted and the work must continue from the first unfinished step
- A teammate in Qoder or Codex finishes what was started in Claude Code
- Several agents split the work and the plan owner needs one place for verification evidence
- You want the decisions on a task stored with the code and discussed in a pull request
- The task runs as a list of steps and each step is closed with a mark after verification
Not a fit when
- You need an answer to a question or a code walkthrough: such requests go directly and never enter a plan
- The fix is small, such as a rename or a typo: a plan is overkill
- The script is throwaway or a quick try: there is no point in keeping history
Example request
~go plan add input validation to the signup formLimitations
Python 3.11 or newer is required, and the installer checks it before downloading. In Codex, Claude Code and Qoder the workflow was verified in real sessions, while for Copilot, Cursor, Antigravity and Trae only the installation is confirmed.
The installer writes instructions and skills into the home directories of the chosen tool. On Windows, the `install.ps1` PowerShell script replaces `curl`.
In non-interactive runs such as `-p` and `--print`, the agent's question ends the turn and the request has to be sent again.
How to disable. Keep the `.sopify/` directory: it holds your plans and history. Remove the block between `<!-- BEGIN SOPIFY MANAGED BLOCK -->` and `<!-- END SOPIFY MANAGED BLOCK -->`, markers included, from the instruction file: `~/.claude/CLAUDE.md` for Claude Code, `~/.codex/AGENTS.md` for Codex, `~/.qoder/AGENTS.md` for Qoder, `.github/copilot-instructions.md` for Copilot. The `skills/sopify/` and `sopify/` folders in the tool home directory can go too.
FAQ
How do I install Sopify in Claude Code?
Run `curl -fsSL https://github.com/evidentloop/sopify/releases/latest/download/install.sh | bash -s -- --target claude:en-US`. The `claude:zh-CN` variant installs the Chinese version, and Python 3.11 or newer is required.
What lands in the repository?
The `.sopify/sopify.json` file and the `.sopify/plan/` and `.sopify/history/` directories are committed. The `.sopify/state/` directory with the pointer to the active plan stays local: the installer excludes it through `.git/info/exclude`.
When is a plan created?
Only on `~go plan`, on a request to continue work in progress, or when `~go` decides a plan is needed. Questions, code reading and small fixes go without a plan.
Which tools does Sopify work in?
The workflow was verified in real sessions of Codex, Claude Code and Qoder. For Copilot, Cursor, Antigravity and Trae the installation is confirmed, while evidence from real sessions varies.
How do I close a finished plan?
With `~go finalize` once all tasks are done. The plan moves to `.sopify/history/` and the verification evidence stays next to it.
Discussion
No questions or reviews yet.
Related
SkillsThe /improve skill: a strong model audits a repository and writes self-contained fix plans, and you hand the execution to a cheaper model
A Claude Code skill that drives a hard task through explore, plan, execute and reflect, writing every decision to disk.
A workflow for coding agents: a plan you approve, evidence-based verification and a handoff of state between sessions.
A planning extension for the Pi coding agent: reviewed Markdown plans first, and code changes only after your approval
Codely team skills that run agent work in phases: a plan in a file or GitHub issues, review after each phase and a clean commit
An agreed plan, a check after each wave of edits and a handover to the next session for Claude Code and Codex.
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| Sopify (this card) | no | 369 | |
| Spec Kit | no | 149,434 | GitHub's spec-driven toolkit with the `specify` CLI, while Sopify is built around a plan, verification evidence and resuming with `~go` |
| OpenSpec | no | 74,686 | leads from a proposal and requirements to tasks and code, while here the focus is moving work between agents and an archive of closed plans |
README in short
The text starts with a one-command install and a first plan, then covers where it pays off and its limits, a command cheat sheet, a table of supported tools and a real Codex example. It ends with links to docs on how it works, installation, the Orca roles and onboarding new tools.
README badge
Original SKILL.md
--- name: analyze description: Analyze phase entry. Aggregates scoring, follow-up, and scope-check rules; loads references/assets/scripts on demand. --- # Analyze (Entry) > Read the evidence first, then separate the objective, deliverable, and candidate path. ## When to activate - Entering the analysis phase (`workflow` / `plan_only`). - Need requirement scoring, clarification, or complexity routing. ## Execution skeleton 1. Load `references/analyze-rules.md` first. 2. Read user-supplied material and relevant code, separate the objective, deliverable, and candidate path, then score completeness. 3. If the score is below threshold, follow `auto_decide`: - `false`: ask follow-up questions with `assets/question-output.md`. - `true`: state explicit assumptions, then continue. 4. After the score passes, hand off confirmed success criteria, scope, evidence, and unknowns that could still affect the result. 5. Render the phase summary with `assets/success-output.md`. ## Resource navigation - Long rules: `references/analyze-rules.md` - Shared writing standards: `../references/shared-writing-dna.md` (apply to all output) - Output contract: `../references/output-contract.md` (required sections, conditional enhancement, self-check) - Follow-up template: `assets/question-output.md` - Success template: `assets/success-output.md` - Deterministic scoring script: `scripts/score_requirement.py` ## Deterministic logic first Use the script when the score must be auditable: ```bash python3 skills/en/skills/sopify/analyze/scripts/score_requirement.py \ --goal-clarity 2 \