specrails — Claude Code plugin: what it does, how to install
specrails installs an agent workflow into your repository, setting up architect, developer, and reviewer stages for Claude Code.
MIT ★ 9 updated 2026-10-24
/plugin marketplace add anthropics/claude-plugins-community
/plugin install specrails@claude-communityWhat specrails does
The specrails plugin introduces an agent-driven development workflow into a codebase. It sets up the implement and retry entry points, provider configurations, and OpenSpec skills. The runtime defines roles like architect, developer, and reviewer internally, so no external role definition files are added to the project.
The execution engine handles deterministic verification receipts, durable recovery, and role-specific models compatible with the OpenAI API. Specrails Core does not push commits or publish pull requests directly; it prepares a reviewed candidate changeset for a host application.
What it helps with
- execute feature tickets through specialized AI roles
- resume interrupted implementation runs using the retry entry point
- run programmatic agents on OpenAI-compatible local models
- verify phase outputs before handing candidate code to the host
Example prompts for the agent
- Run the ticket implementation using the
implemententry point - Resume the failed workflow step with
retry - Inspect the current implementation phase with
specrails-core pipeline
How to install specrails
Plugin marketplace
/plugin marketplace add anthropics/claude-plugins-community/plugin install specrails@claude-communityStep by step, with commands for each platform: Claude Code plugins: marketplaces, installing, and building your own
Questions about specrails
How to install specrails in Claude Code?
First, add the marketplace registry with /plugin marketplace add anthropics/claude-plugins-community. Then install the package by running /plugin install specrails@claude-community.
Does specrails push commits and pull requests automatically?
No. The core engine only prepares a reviewed candidate for the host environment. Git worktrees, commits, and pull requests are managed by host tools such as Specrails Desktop.
Can specrails work with local LLMs?
Yes. Its agent runtime supports per-role providers and connects to local models through an OpenAI-compatible API.
Source text
specrails-plugin · MIT
# specrails-core [](https://github.com/fjpulidop/specrails-core/actions/workflows/ci.yml) [](https://www.npmjs.com/package/specrails-core) [](LICENSE) The agent workflow engine behind [Specrails Desktop](https://github.com/fjpulidop/specrails-desktop). Core does two things: 1. **Installs** the specrails workflow into a project for Claude Code, Codex CLI, Gemini CLI or Kimi Code: the `implement` and `retry` entry points (`implement` runs one ticket or several tickets as one aggregate run), provider settings and the official OpenSpec skills. The architect, developer and reviewer roles are defined by the runtime itself; no role file is installed. 2. **Executes** Desktop-owned workflow definitions through the [programmatic agent runtime](docs/agent-runtime.md): independent agent and implementation operations, per-role providers, deterministic verification, durable recovery and OpenAI-compatible local models. Desktop owns workflow composition and editable loop agents; Core retains the old builtin for saved runs. Desktop bundles a pinned Core package and owns everything around it: worktrees, commits, pull requests, backlog and the user interface. Core never ships code on its own; it prepares a reviewed candidate for the host. ## Commands ```text specrails-core init Install specrails into a repository specrails-core install-framework Materialize the versioned framework (offline) specrails-core swap-current Point framework/current at a version (offline) specrails-core assemble Link the framework into a project workspace (offline) specrails-core pipeline Inspect and verify durable implementation phases specrails-core runtime Run, inspect and resume programmatic agents ``` Desktop calls `init` and the offline framework lifecycle, then launches `dist/agent-runtime/cli.js` directly. The machine-readable contract between both lives in [`integration-contract.json`](integration-contract.json); the runtime configuration schema is [`schemas/agent-runtime.schema.json`](schemas/agent-runtime.schema.json). ## Layout ```text bin/specrails-core.mjs npm bin shim → dist/installer/cli.js src/installer/ CLI, framework lifecycle, provider rendering src/pipeline/ pipeline journal, gates and verification receipts src/agent-runtime/ programmatic runtime: graph, executors, compact loop, recovery templates/ workflow commands, provider settings, Kimi runner schemas/ runtime configuration schema and fixtures integration-contract.json Desktop ⇄ Core contract ``` ## Develop Requires Node.js 20.19.0+ and Git. ```sh npm ci npm test # build + typecheck + vitest npm run ci # full local CI: typecheck, script tests, coverage, package check ``` See [CONTRIBUTING.md](CONTRIBUTING.md) and [CI and publishing](docs/ci-cd.md). ## License MIT
© fjpulidop · MIT · fjpulidop/specrails-core