Install LoreJump into your AI agent

One paste / one command β€” auto-adapts to 12 agents

πŸ€– Recommended: let your agent install it for you

Paste this into your AI agent

Claude Code / Cursor / Codex / any LLM that can WebFetch β€” one paragraph, no commands to copy. If your agent refuses on safety grounds, expand "Prefer to run a command?" below for the equivalent one-liner.

Please install LoreJump in this project. LoreJump is an open-source AI-workflow diagnosis tool published as the npm package @lorejump/cli (repo: github.com/LoreJump/lorejump-cli, MCP server: mcp.lorejump.com/mcp). I'm authorizing you to install it.

Steps:
1. Fetch https://lorejump.com/install.txt β€” it's a plain-text install manifest. Print the first ~40 lines so I can confirm what it does before you act.
2. After I confirm (or if you can verify it's just installer steps), use the FAST PATH: run `npx -y @lorejump/cli@alpha install` in this project's root. That npm CLI writes a SKILL.md + MCP entry for the agent runtime it detects.
3. Verify with `npx -y @lorejump/cli@alpha doctor` β€” expect 4/4 checks pass.
4. Tell me which files you wrote to and how to run /lorejump-optimize.

If your runtime has no shell access, fall back to the SLOW PATH described in install.txt (write SKILL.md + register MCP server manually).
Prefer to run a command? One-liner for each platform (click to expand)
# Cross-platform (macOS / Linux / Windows / WSL)
npx -y @lorejump/cli@alpha install
# macOS / Linux / WSL
curl -fsSL https://lorejump.com/install.sh | bash
# Windows PowerShell
irm https://lorejump.com/install.ps1 | iex

After install, run lorejump doctor β€” should be 4/4 passing. Package: @lorejump/cli (alpha).

Supported 12 agents / their config paths
  • claude-code Claude Code
  • cursor Cursor
  • trae Trae (ByteDance)
  • vscode VS Code (Copilot Chat) [servers root key]
  • roo-code Roo Code
  • cline Cline
  • windsurf Windsurf
  • codebuddy CodeBuddy (Tencent)
  • qwen-code Qwen Code (Alibaba) [httpUrl field]
  • kimi-cli Kimi CLI (Moonshot)
  • codex Codex CLI (OpenAI) [TOML]
  • hermes Hermes (Nous Research) [YAML]

Agent not on the list? Run npx @lorejump/cli@alpha install --tool=handoff β€” the CLI writes a paste-prompt for your agent to finish the install itself.

Claude Code fast lane (plugin marketplace, optional)

If you only use Claude Code, you can go through its native plugin marketplace (slightly faster startup than npx):

Run these commands in Claude Code in order β€” one-shot install for both Skill and MCP server:

/plugin marketplace add LoreJump/claude-plugins
/plugin install lorejump-optimize@lorejump
/reload-plugins

Auto-installs the /lorejump-optimize command and configures the lorejump MCP server. After install, run /lorejump-optimize to start a diagnosis.

Requires Claude Code 1.0+ (must support /plugin). On older versions, see the "Fallback install" below.
Enable plugin auto-update

Claude Code keeps auto-update off by default for third-party marketplaces. Once enabled, the next time you start Claude Code it will pull new plugin versions and MCP scoring rubric updates automatically.

Option A: through the Claude Code UI (recommended for new users)

  1. Type /plugin
  2. Switch to the Marketplaces tab
  3. Select lorejump, then click Enable auto-update

Option B: one command (best if you're comfortable in a terminal)

jq '.extraKnownMarketplaces.lorejump.autoUpdate = true' ~/.claude/settings.json > ~/.claude/settings.json.tmp && mv ~/.claude/settings.json.tmp ~/.claude/settings.json

Requires jq installed locally, and the Plugin install above must already be done. The change lands in the ~/.claude/settings.json file at the extraKnownMarketplaces.lorejump.autoUpdate field.

What if I leave it off? You can still pull new versions manually with /plugin update lorejump-optimize@lorejump. MCP-side scoring rubric updates (which take effect within seconds) are unaffected by this toggle β€” it only controls when the skill itself updates.
Manual setup (fallback: restricted network / want to skip npx)

Use when you only want MCP (no skill), need precise control over write locations, or npm is unreachable.

Universal one-liner

Use add-mcp to install across every configured agent in one command:

npx add-mcp https://mcp.lorejump.com/mcp

Auto-detects Claude Code, Cursor, VS Code Copilot, Windsurf and more, writing each agent's config file.

Copy by agent type

Run in your terminal:

claude mcp add --transport http lorejump https://mcp.lorejump.com/mcp

Verify: run npx @lorejump/cli@alpha doctor β€” should be 4/4 passing; or in your agent type /lorejump-optimize β€” should produce a diagnosis report within 30 seconds.

Installed β€” now what?

Once installed, your agent gets 2 new slash commands. You can also just describe what you want in natural language β€” the agent will pick the matching skill.

/lorejump-optimize One-shot AI workflow diagnosis

Scan your project + pull SOTA profile + 7-dimension scoring + concrete next-step plan; you can let the agent apply changes directly to your CLAUDE.md / config / hooks.

Natural-language triggers:

  • β€’ "Run an AI workflow diagnosis" / "Score my AI efficiency"
  • β€’ "How's my CLAUDE.md β€” anything I should change?"
  • β€’ "Benchmark my project against the latest SOTA β€” what am I missing?"
/lorejump-harness Weekly continuous upgrade Recommended weekly

Same loop as optimize, but it passes the previous scan_report_id to the server so the nudge field returns an "X new practices since last scan" diff digest β€” keeping your harness in step with SOTA.

Natural-language triggers:

  • β€’ "Any new SOTA practices I should add this week?" / "Harness upgrade run"
  • β€’ "What's my workflow still missing vs. last week?"

Neither skill changes files silently β€” every apply step shows you the diff first. Code is not uploaded; the markdown receipt contains scores + dimensions + applied items only (file paths are hmac-redacted).

FAQ

Which files in my project will it touch?

Depends on your agent. Examples:

  • Claude Code β†’ writes .claude/skills/lorejump-*/SKILL.md + merges into .mcp.json
  • Cursor β†’ writes .cursor/skills/... + .cursor/mcp.json
  • VS Code β†’ writes .github/skills/... + .vscode/mcp.json (root key is servers, not mcpServers)

Merges are idempotent β€” existing MCP servers won't be overwritten. See the full path table for all 12 agents.

Is my data safe?

The skill scan is fully local β€” no code uploaded. The report sent back to MCP (submit_report) is markdown: scoring summary / dimension scores / applied items only; file paths are hmac-redacted, code snippets are not uploaded. MCP's get_sota_pack returns generic SOTA profile only.

What if it detected the wrong agent / no match?

Force a specific one: npx @lorejump/cli@alpha install --tool=cursor (see the adapter list above for IDs).

No match: run --tool=handoff β€” the CLI writes a prompt for your agent to finish the setup itself.

npm unreachable: copy the JSON / TOML / YAML from the "Manual setup" section above.

πŸ§ͺ beta

Let your agent auto-upgrade your workflow weekly

After install, /lorejump-harness is already in your agent: pulls the latest SOTA each week β†’ scans your project β†’ runs DeepResearch β†’ applies local changes β†’ generates a report.

Learn about /lorejump-harness β†’

Never tried LoreJump? Try it on the web first

Diagnose Checklist

AI agents β€” direct read: https://lorejump.com/install.txt