vibe-coding-analytics

Initialize, audit, and evolve project harnesses for modern AI coding agents.

vibe-coding-analytics treats vibe coding as an engineering loop — clear instructions, scoped memory, repeatable checks, reviewer roles, reusable skills, and recurring improvement. It ships as a single npm package with a zero-config CLI and first-class templates for Claude Code and Codex.

✨ Features

• Audit a project for AI coding readiness across agents, memory, skills, tests, CI, and deploy hooks.
• Scaffold baseline harness files (AGENTS.md, CLAUDE.md, skills, slash commands, reviewer agents).
• Evolve a concrete improvement plan: maps detected analytics gaps to promotion targets (tests, validators, CI, skills, rules) and surfaces recent git fix hotspots as regression-test candidates.
• Monorepo-aware — git submodules, npm workspaces, polyglot stacks, and fractal per-directory docs are detected automatically.
• Read-only by default — every command is safe to dry-run; pass --write to materialize changes.

📦 Installation

Run without installing:

npx vibe-coding-analytics analytics

Or install globally:

npm install -g vibe-coding-analytics

Requires Node.js >= 18.

🚀 Usage

# One-off, no install
npx vibe-coding-analytics analytics
npx vibe-coding-analytics init --write
npx vibe-coding-analytics evolve --write

vca is also exposed as a CLI alias. With npx, use --package so npm installs the vibe-coding-analytics package and then runs its vca binary:

npx --package vibe-coding-analytics vca analytics

With a global install:

vibe-coding-analytics analytics
vibe-coding-analytics init --write
vibe-coding-analytics evolve --write

# or the short alias
vca analytics
vca init --write
vca evolve --write

Commands are read-only by default. Add --write to create missing harness files.

The npm package and CLI are the same package — there is no separate CLI package to publish or install.

Commands

| Command | Purpose | | --- | --- | | analytics | Audit the current project for AI coding readiness. | | init | Propose or create baseline harness files. | | evolve | Propose or create self-evolution loop files for repeated improvements. |

Integrations

Claude Code Slash Commands

The generated project templates support:

/analytics
/init
/evolve

The evolve command is designed to pair with local loops:

/loop 30m /evolve

Use it to extract repeated user corrections, failed checks, review feedback, and repeated command sequences into durable tests, validators, rules, commands, skills, or reviewer agents.

Codex Skill

This repository includes a distributable skill at:

skills/vibe-coding-analytics/SKILL.md

The skill triggers when initializing a new project, auditing an existing repository for AI coding readiness, adding agent rules, or improving a project's vibe coding harness.

Codex Plugin

The repository includes a Codex plugin manifest:

.codex-plugin/plugin.json

It points Codex at the included skills directory and can be used as the base for marketplace submission.

What It Checks

• AGENTS.md, CLAUDE.md, Cursor rules, and Copilot instructions
• Project memory such as PRDs, constraints, patterns, known issues, or fractal per-directory CLAUDE.md
• Reusable skills, slash commands, specialist reviewers (.claude/agents/, reviewer skills, or plugin agents)
• Tests, typecheck, lint, architecture validators, and CI
• Rule-to-sensor coverage: prose rules in CLAUDE.md/AGENTS.md backed by computational sensors (tests, lint, validators), not prose-only
• Steering loop: numbered rules (Rule N) in CLAUDE.md/AGENTS.md that grow after each bug fix — the rising count is the feedback-loop heartbeat
• Failure observability: monitoring, alerting, or error counters so critical-path failures surface instead of failing silently
• Deploy and post-deploy verification hooks
• Cross-session memory: ADR decisions, agent memory, or a decisions log
• Self-evolution loops that promote repeated work into durable harness assets
• Depth hints on passing checks (test-file / instruction-line / skill counts) so a stub is distinguishable from a mature project at the same score
• False-safety warnings when checks are partially present (tests without CI, agent rules without enforcement, no single validation command)

Monorepo & Non-Standard Conventions

The analyzer is aware of common repository shapes and does not assume a flat single-package Node project:

• Git submodules — when .gitmodules is present, every check runs against the root plus each submodule root, so tests, scripts, and harness files living one level down are still detected.
• npm workspaces — detected from the root package.json workspaces field.
• Polyglot stacks — package.json scripts are merged from every subpackage; Go (go.mod), Flutter (pubspec.yaml), and TypeScript (tsconfig.json) are recognized as typecheck signals; test files are detected by convention (*_test.go, *_test.dart, *.test.{js,ts}, test_*.py, …).
• Fractal docs — two or more CLAUDE.md/AGENTS.md files count as project memory, matching the per-directory documentation pattern.
• Architecture sensors — any scripts/*validate*, *verify*, *check*, or *lint* file counts, not just scripts/validate-architecture.

The report prints the detected Project shape and how many roots were scanned.

🤝 Contributing

Contributions are welcome! Please open an issue or submit a pull request.

📄 License

MIT © 2026 kx-code