Skip to content
项目

caliber-ai-org/ai-setup has 1,320 GitHub stars and 130 forks. GitHot measured +1 stars in 24 hours, +5 in 7 days, and +11 in 15 days. Last checked 2026-11-22 16:27 UTC. Last push 2026-09-24.

这些数字如何统计

实测星标增长

24小时
+1
7天
+5
15天
+11
1个月
正在收集历史

各周期内的星标净变化,周期相互重叠,并非每日新增。缺失的测量保持空白。星标代表关注度,不代表软件质量。

关于

用一条命令持续同步你的 AI 配置。为 Claude Code、Cursor 和 Codex 提供适配代码库的智能体技能、MCP 及配置文件。

开发工具

README.md

原文

Caliber

The open-source context layer for coding agents.
Claude Code · Cursor · Codex · OpenCode · GitHub Copilot

npm version MIT license Node version

01 Fresh: context that follows the code. 02 Shared: write a skill once, every agent has it. 03 Lean: compact without summarizing.

Start

npx @rely-ai/caliber bootstrap

Then, in Claude Code or the Cursor CLI, run /setup-caliber. Your agent writes CLAUDE.md, Cursor rules, AGENTS.md and Copilot instructions, shows you the diff, and installs the hooks that keep them current. No agent handy? caliber init does the same as a wizard.

Node 20+. Runs on your Claude Code or Cursor seat, or your own Anthropic, OpenAI, MiniMax or Vertex key. Bootstrap, scoring and sync run fully local.

Fresh

A pre-commit hook reads your diff and refreshes the context files your agents read. caliber score checks them against the repo, with no LLM: do the paths exist, are the commands real, has anything drifted.

caliber score --compare main

Shared

Write a skill, rule, plugin or MCP server in one agent and caliber sync puts it in every other one. It's deterministic, degrades where an agent lacks a feature (a skill becomes a Copilot instruction file), and never overwrites a file you edited by hand.

caliber sync --status

Lean: Jev compaction new

Built-in /compact turns the session into a lossy paragraph. Jev scores each tool call, drops the stale ones and keeps every message verbatim.

/compact summarizes, and summaries forget the file path, the error and the constraint you gave an hour ago. TypeSafe's Jev scores every tool call, drops the stale ones, and keeps everything else byte-for-byte.

Claude Code: a plugin that replaces /compact.

export CLAUDE_CODE_ENABLE_FUNCTION_HOOKS=1
export AI_GATEWAY_API_KEY=...        # or TYPESAFE_API_KEY
claude plugin marketplace add caliber-ai-org/ai-setup
claude plugin install caliber-jev-compaction@caliber

Cursor, Codex and any other agent: the same scoring from the CLI.

caliber compact --provider cursor  --transcript <jsonl>
caliber compact --provider generic --transcript ./session.json --write

Bring your own Vercel AI Gateway or TypeSafe key. If Jev is unavailable, Claude Code falls back to its built-in compaction. Setup and verification: FUNCTIONAL_CHECKLIST.md.

Commands

Command
caliber bootstrap / init Set up (agent skill or CLI wizard)
caliber score Audit config quality, no LLM
caliber refresh Update configs from recent changes
caliber sync Mirror skills, rules, plugins and MCP across agents
caliber compact Jev compaction report (--write for generic transcripts)
caliber plugin install Install the Jev plugin into this project
caliber learn Learn patterns from your sessions into CALIBER_LEARNINGS.md
caliber hooks Manage refresh and sync hooks
caliber config Choose provider, key and model
caliber undo / uninstall Revert changes / remove everything Caliber added

Every write is shown as a diff first and backed up to .caliber/backups/. Anonymous usage analytics (command names, never code) can be turned off with CALIBER_TELEMETRY_DISABLED=1.

Contributing

git clone https://github.com/caliber-ai-org/ai-setup.git && cd ai-setup
npm install && npm run test

See CONTRIBUTING.md. MIT licensed.

探索相关项目

加载中…