Caliber
A CLI that generates and maintains AI context files, CLAUDE.md, Cursor rules, AGENTS.md and Copilot instructions, keeping them accurate as code changes
Medium risk
We rate an entry medium when the tool runs code, makes network calls or reads project files. Check what exactly it does before installing.
Why this level
- Installs git and session-end hooks that run automatically
- Sends a project summary to a model provider and collects anonymous analytics by default
- Reads project files and writes configs, and handles provider keys
Install
In your terminal, with SkillFoxx CLI
npx skillfoxx add cli/caliberDetects the agents on your machine, checks the risk and pins the version.
Other ways to install
Install the tool
npm install -g @rely-ai/caliberInstall Caliber with one command: npx @rely-ai/caliber bootstrap. Then in a Claude Code or Cursor CLI session run /setup-caliber so the agent detects the stack and generates the configs. Without an agent use caliber init.
Other ways from the author
npx @rely-ai/caliber bootstrapOne-time setup that installs the setup-caliber skill for the agent. Needs Node.js 20 or newer.
This is third-party code. Review the repository files before installing.
What it does
Caliber generates and updates context files for coding agents so they do not go stale after a refactor. A single npx @rely-ai/caliber bootstrap installs the /setup-caliber skill, then the agent detects your stack, generates configs for every platform your team uses and installs hooks. The caliber score command gives a deterministic quality score with no model call: it cross-references paths and code blocks against your real project tree. Pre-commit and session-end hooks refresh the docs from code changes. Every edit is shown as a diff, originals are backed up, and caliber undo reverts them.
Who it is for. For developers and teams who keep context files for Claude Code, Cursor, Codex, OpenCode and Copilot up to date.
Good fit when
- The agent's context files are stale after a refactor
- You need one project description shared across several agents
- You want to score CLAUDE.md and rules quality without a model call
Not a fit when
- Keeping one short file by hand is enough
- You have no Node.js 20 or a suitable model provider for generation
Example request
Check and refresh my CLAUDE.md and Cursor rules against the current codeLimitations
It needs Node.js 20 or newer. Bootstrap and scoring run locally with no model, while generation calls a provider: a Claude Code or Cursor subscription, or an Anthropic, OpenAI, MiniMax or Vertex AI key. Some providers are not directly available from Russia, so availability depends on the option you pick. Generation sends out a project summary (languages, structure, dependencies, not source code), and anonymous analytics are collected by default via PostHog, which you can turn off. Keys are stored in ~/.caliber/config.json with restricted permissions.
How to disable. Remove Caliber's resources from the project with caliber uninstall, revert edits with caliber undo, and drop the hooks with caliber hooks --remove.
Security check
- Installs git and session-end hooks that run automatically
- Sends a project summary to a model provider and collects anonymous analytics by default
- Reads project files and writes configs, and handles provider keys
README in short
The README presents Caliber as a tool that generates and maintains context files for coding agents and keeps them in sync across Claude Code, Cursor, Codex, OpenCode and Copilot. Installation is npx @rely-ai/caliber bootstrap, after which the agent runs the /setup-caliber skill, or you use the caliber init wizard without an agent. The caliber score check is deterministic and calls no model, cross-referencing configs against the real project tree across six categories. Edits are shown as diffs, originals are backed up, and there is undo, dry-run and clean uninstall. Generation works with a Claude Code or Cursor subscription or with Anthropic, OpenAI, MiniMax and Vertex AI keys. MIT licensed.
FAQ
Does my code leave the machine?
Scoring is fully local. Generation sends only a project summary, not source code, to the provider you choose.
Do I need an API key?
Not for bootstrap and scoring. Generation works with a Claude Code or Cursor subscription or your own Anthropic, OpenAI, MiniMax or Vertex AI key.
Related
A skills library that gives coding agents a development process: brainstorming, planning, TDD, subagents and code review
Skills for real engineers by Matt Pocock
Skills For Real Engineers
Small composable skills for engineering with agents: plan grilling, TDD, bug diagnosis, code review and architecture
GitHub toolkit for spec-driven development: the specify CLI adds agent commands and skills to a project, from principles to implementation
Reference MCP servers
Model Context Protocol servers
Official reference MCP servers: Filesystem, Fetch, Git, Memory, Sequential Thinking, Time and Everything