Grades eval variant outputs for run-skill-eval. Reads metrics.json from @spine:run-skill-eval:runner, evaluates assertions, writes grading.json.
Designs eval test cases for run-skill-eval. Generates quality evals and trigger evals for a single skill/agent/instruction unit.
Generates optimized skill variants for run-skill-eval. One dispatch per optimization angle. Outputs a single variant file.
Craft review for run-skill-eval. Audits skill/agent files against authoring standards. Outputs findings that feed into variation generation.
Runs calibration + eval execution + metrics extraction for run-skill-eval. Dispatched per eval unit. Outputs structured metrics, not raw stream-json.
Executes bash commands
Hook triggers when Bash tool is used
Modifies files
Hook triggers on file write and edit operations
Uses power tools
Uses Bash, Write, or Edit tools
Own this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimOwn this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimBased on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
Same skills, same workflow, every provider.
Spine is a shared AI coding setup for Cursor, Claude Code, Codex, and OpenCode. It gives each tool the same workflow skills, subagents, guardrails, and MCP defaults so you can move between providers without rebuilding your operating model.
If it's worth changing, it's worth planning.
Install and sign in to the provider tools you want Spine to configure first. Spine does not install provider CLIs for you.
| Provider | CLI | Host | Skills | Subagents | Envoy | Notes |
|---|---|---|---|---|---|---|
| Claude Code | claude | Full | Full | Full | Default | Strong planning/review. Opus escalation for high-risk work. |
| Codex | codex | Full | Full | Full | Default | Best current long-horizon coding signal (DeepSWE/GPT-5.5). |
| Cursor | cursor-agent | Full | Full | Partial¹ | Default | Best IDE integration. Monthly cap. |
| OpenCode | opencode | Full | Full | Full | Availability-gated | Multi-model gateway. Strong for cost/diversity; weaker DeepSWE coding signal. |
¹ Legacy plans ignore subagent model config.
Envoy dispatches cross-provider perspectives during design and review. Default: 3 core providers (claude, codex, cursor) + availability-gated opencode. Override with SPINE_ENVOY_PROVIDERS.
For tier-to-model mappings, pricing, and benchmarks see docs/model-selection.md.
Provider Prerequisites
Cursor works with the editor alone. Install the CLI only if you want terminal access to agent commands:
curl https://cursor.com/install -fsS | bash
echo 'export PATH="$HOME/.local/bin:$PATH"' >> ~/.zshenv
source ~/.zshenv
agent login
brew install codex
codex login
Spine's envoy skill can invoke Claude Code from within a Codex session. On macOS, Codex runs shell commands inside a Seatbelt sandbox that blocks macOS Keychain access. Claude Code stores OAuth credentials in Keychain, so claude auth status fails under the default sandbox.
To enable cross-provider invocation, relax the sandbox in ~/.codex/config.toml:
sandbox_mode = "danger-full-access"
approval_policy = "on-request" # keeps human-in-the-loop approval
Or per-session via CLI flag:
codex -s danger-full-access
| Setting | What it does |
|---|---|
sandbox_mode | read-only (default), workspace-write, or danger-full-access (disables Seatbelt) |
approval_policy | untrusted (ask every command), on-failure, on-request (auto within boundaries), never |
danger-full-access disables the Seatbelt sandbox entirely — shell commands run with full system access. Keep approval_policy = "on-request" so Codex still asks before running unexpected commands.
This is only required for cross-provider envoy (Codex calling Claude). Standard Codex usage and Claude calling Codex work without this change.
curl -fsSL https://claude.ai/install.sh | bash
claude auth login
brew install anomalyco/tap/opencode
opencode providers
OpenCode is a multi-model gateway supporting GLM, MiniMax, and many others. Spine uses two pricing tiers:
opencode-go/ prefix) — included in subscription, free at marginopencode/ prefix with -free suffix) — zero cost Zen modelsSee opencode.ai/docs/models for the full model catalog.
curl -fsSL https://raw.githubusercontent.com/kenoxa/spine/main/install.sh | bash
~/.cursor/claude CLI on PATHcodex CLI on PATHopencode CLI on PATHFor Claude Code, the installer also attempts to install the Spine plugin.
Autonomous mode — skills dispatch subagents that read, write, and run commands without approval prompts. Other modes interrupt this with constant approval requests.
npx claudepluginhub kenoxa/spine --plugin spineUltra-compressed communication mode. Cuts 65% of output tokens (measured) while keeping full technical accuracy by speaking like a caveman.
Multi-model consensus engine integrating OpenAI Codex CLI, Gemini CLI, and Claude CLI for collaborative code review and problem-solving.
Memory compression system for Claude Code - persist context across sessions
Comprehensive UI/UX design plugin for mobile (iOS, Android, React Native) and web applications with design systems, accessibility, and modern patterns
Unified capability management center for Skills, Agents, and Commands.
45% cost reduction measured. Cache expiry prevention, SubTask auto-delegation, zero-cost context restoration, real-time cost dashboard. The only Claude Code plugin built from CC source analysis.