vibelint — Summary

vibelint is a Node.js CLI linter that inspects local Claude Code session data and static repository configuration to detect quality problems, model-choice mismatches, and cost signals in AI-assisted development workflows. It operates post-session (read-only analysis of ~/.claude transcripts and project config) and never modifies code or Claude settings. The tool is invoked via npx vibelint with no install requirement, making it trivially embeddable in CI pipelines. It reports configuration quality, permission-setting issues, command coverage, conversation efficiency, and cross-session trends. Unlike runtime guardrails, vibelint is strictly a reporting/audit tool: it has no hooks, no policies, and no blocking mechanism.

Compared to seeds: closest to the "lint" pole occupied by spec-kit's 18 hooks that cross-validate context files, but vibelint intervenes after sessions complete rather than before/during tool calls. It resembles claude-conductor in that its value is entirely in markdown/context-file hygiene, but vibelint drives that signal programmatically from session logs rather than shipping template files.

vibelint — Overview

Origin

Repo: l-mendez/vibelint (github.com/l-mendez/vibelint). Published March 2026, TypeScript, MIT. 2 stars. The repo description: "The linter for vibecoding." Small personal project by a single author.

Philosophy

The tool is named after "vibecoding" — a colloquial term for AI-assisted development driven by intuition and rapid iteration rather than formal engineering discipline. vibelint applies traditional software-linting thinking to the AI session layer: you can lint JavaScript files for quality; why not lint the conversations your agent has with your codebase?

The README frames the problem plainly:

"vibelint inspects your current repository plus local Claude Code session data and reports configuration issues, session quality patterns, and model-usage signals."

Key belief: Claude Code session logs in ~/.claude carry enough signal (model choices, command invocations, output patterns) to assess whether vibe-coded sessions are actually productive or just expensive and noisy.

Manifesto-Style Quotes

From README:

"If no Claude Code sessions are available, vibelint can still run static checks with --static."

(implying the tool is designed to degrade gracefully — static linting alone has value without session data)

Anti-patterns Targeted

• Model-choice mismatch (using a large model when a small one would suffice)
• Stale/incorrect project config visible to agents
• Conversations that have low output-to-token ratios
• Missing command coverage in CLAUDE.md

vibelint — Architecture

Distribution

• npm package, published as vibelint
• No install required: npx vibelint downloads ~minimal bundle on first run
• Global install: npm install -g vibelint

Install Complexity

One-liner (npx vibelint).

Directory Structure (repo)

vibelint/
  src/
    index.ts          CLI entry point
    cli.ts            commander.js setup + flags
    core/             check implementations
    utils/            helpers
    __tests__/
  tests/
  package.json
  tsconfig.json
  vitest.config.ts

Required Runtime

• Node.js 20 or newer
• A git repository to scan
• ~/.claude/ directory for session-based analysis (optional — --static skips)

Target AI Tools

Claude Code exclusively (reads ~/.claude session and config format).

Config Files

None — vibelint reads existing project files (package.json, CLAUDE.md, etc.) and Claude's own session store. It writes no config.

Key Data Sources

• ~/.claude/ — session transcripts, permission settings
• Project root CLAUDE.md, package.json, Makefile
• .claude/ directory (local Claude Code settings)

vibelint — Components

CLI Binary

vibelint — single binary, no subcommands. All behavior controlled by flags.

Flags

Check Categories

From README:

Source Modules

• src/cli.ts — createCli() with Commander.js setup
• src/core/ — check implementations (exact file names not publicly exposed)
• src/utils/ — shared helpers

No Hooks, Scripts, or MCP Servers

vibelint ships zero hooks, zero skills, zero MCP tools. It is a pure read/report binary.

vibelint — Uniqueness & Positioning

Differs from Seeds

vibelint occupies a distinct niche from all 11 seed frameworks: it is a post-session audit tool rather than a session-steering framework. The closest seed is spec-kit (which also has 18 hooks for context-file linting and cross-validation), but spec-kit's hooks fire during agent sessions (PreToolUse/PostToolUse) to enforce correctness in real-time. vibelint fires after sessions complete, reading transcript artifacts to score session quality for the developer's benefit — not the agent's. claude-conductor is also context-file hygiene focused but ships markdown templates rather than a linting engine. vibelint is the only tool in the seed set that treats session transcripts as the primary lint target.

Positioning

Signal type: post-session audit / cost reporting Intervention point: none (no blocking) — pure read + report Target user: developer reviewing AI session quality, CI pipeline enforcing minimum quality gates Unique feature: model-choice mismatch detection and cross-session cost trends

Observable Failure Modes

• If Claude Code changes its session file format, vibelint breaks silently (no schema validation shown in source)
• Limited to Claude Code — any team using Cursor/Copilot/Codex gets zero value
• No auto-fix capability (ctxlint in this batch has --fix for broken path references)
• Tiny install base (2 stars) — unclear how actively maintained checks are

Relationship to Other Batch 31 Frameworks

vibelint is the most minimal intervention-point framework in this batch: it intervenes after everything is done. ctxlint is closer to it (both are linters for AI context) but ctxlint runs on context files before/during sessions. All other frameworks (Sponsio, DashClaw, clauder, pi-steering-hooks) intervene before or during agent tool use.