npx @rely-ai/caliber score

Score your AI agent config in 3 seconds. No API key. No changes to your code. Just a score.

Your code stays on your machine. Scoring is 100% local — no LLM calls, no code sent anywhere. Generation uses your own AI subscription (Claude Code, Cursor) or your own API key (Anthropic, OpenAI, Vertex AI). Caliber never sees your code.

Caliber scores, generates, and keeps your AI agent configs in sync with your codebase. It fingerprints your project — languages, frameworks, dependencies, architecture — and produces tailored configs for Claude Code, Cursor, and OpenAI Codex. When your code evolves, Caliber detects the drift and updates your configs to match.

Most repos start with a hand-written CLAUDE.md and nothing else. Here's what Caliber finds — and fixes:

  Before                                    After caliber init
  ──────────────────────────────            ──────────────────────────────

  Agent Config Score    35 / 100            Agent Config Score    94 / 100
  Grade D                                   Grade A

  FILES & SETUP           6 / 25            FILES & SETUP          24 / 25
  QUALITY                12 / 25            QUALITY                22 / 25
  GROUNDING               7 / 20            GROUNDING              19 / 20
  ACCURACY                5 / 15            ACCURACY               13 / 15
  FRESHNESS               5 / 10            FRESHNESS              10 / 10
  BONUS                   0 / 5             BONUS                   5 / 5

Scoring is deterministic — no LLM, no API calls. It cross-references your config files against your actual project filesystem: do referenced paths exist? Are code blocks present? Is there config drift since your last commit?

caliber score --compare main    # See how your branch changed the score

Caliber never overwrites your existing configs without asking. The workflow mirrors code review:

1. Score — read-only audit of your current setup
2. Propose — generate or improve configs, shown as a diff
3. Review — accept, refine via chat, or decline each change
4. Backup — originals saved to .caliber/backups/ before every write
5. Undo — caliber undo restores everything to its previous state

If your existing config scores 95+, Caliber skips full regeneration and applies targeted fixes to the specific checks that are failing.

Caliber is not a one-time setup tool. It's a loop:

          caliber score
              │
              ▼
  ┌──── caliber init ◄────────────────┐
  │     (generate / fix)              │
  │           │                       │
  │           ▼                       │
  │     your code evolves             │
  │     (new deps, renamed files,     │
  │      changed architecture)        │
  │           │                       │
  │           ▼                       │
  └──► caliber refresh ──────────────►┘
       (detect drift, update configs)

Auto-refresh hooks run this loop automatically — on every commit or at the end of each AI coding session.

Claude Code

• CLAUDE.md — Project context, build/test commands, architecture, conventions
• CALIBER_LEARNINGS.md — Patterns learned from your AI coding sessions
• .claude/skills/*/SKILL.md — Reusable skills (OpenSkills format)
• .mcp.json — Auto-discovered MCP server configurations
• .claude/settings.json — Permissions and hooks

Cursor

• .cursor/rules/*.mdc — Modern rules with frontmatter (description, globs, alwaysApply)
• .cursor/skills/*/SKILL.md — Skills for Cursor
• .cursor/mcp.json — MCP server configurations

OpenAI Codex

• AGENTS.md — Project context for Codex
• .agents/skills/*/SKILL.md — Skills for Codex

caliber init --agent claude        # Claude Code only
caliber init --agent cursor        # Cursor only
caliber init --agent codex         # Codex only
caliber init --agent all           # All three
caliber init --agent claude,cursor # Comma-separated

caliber learn install      # Install hooks for Claude Code and Cursor
caliber learn status       # View hook status, event count, and ROI summary
caliber learn finalize     # Manually trigger analysis (auto-runs on session end)
caliber learn remove       # Remove hooks

caliber hooks --install    # Enable refresh hooks
caliber hooks --remove     # Disable refresh hooks

After scoring your project, add a badge to your README:

Copy this markdown and replace 94 with your actual score:

![Caliber Score](https://img.shields.io/badge/caliber-SCORE%2F100-COLOR)

Color guide: brightgreen (90+), green (70-89), yellow (40-69), red (<40).

No API key? No problem. Caliber works with your existing AI tool subscription:

Override the model for any provider: export CALIBER_MODEL=<model-name> or use caliber config.

Caliber uses a two-tier model system — lightweight tasks (classification, scoring) auto-use a faster model, while heavy tasks (generation, refinement) use the default. This keeps costs low and speed high.

Configuration is stored in ~/.caliber/config.json with restricted permissions (0600). API keys are never written to project files.

# Custom region
export VERTEX_PROJECT_ID=my-gcp-project
export VERTEX_REGION=europe-west1

# Service account credentials (inline JSON)
export VERTEX_PROJECT_ID=my-gcp-project
export VERTEX_SA_CREDENTIALS='{"type":"service_account",...}'

# Service account credentials (file path)
export VERTEX_PROJECT_ID=my-gcp-project
export GOOGLE_APPLICATION_CREDENTIALS=/path/to/service-account.json

• Node.js >= 20
• One LLM provider: your Claude Code or Cursor subscription (no API key), or an API key for Anthropic / OpenAI / Vertex AI

See CONTRIBUTING.md for detailed guidelines.

git clone https://github.com/rely-ai-org/caliber.git
cd caliber
npm install
npm run dev      # Watch mode
npm run test     # Run tests
npm run build    # Compile

Uses conventional commits — feat: for features, fix: for bug fixes.

MIT