菜单
Judgment linter for vibe-coded output — reads the energy of the code, not just correctness. Use when the user says "vibe check", "check this vibe code", "does this hold up", "sanity check this AI code", or after a fast generation session before committing.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Review what agents accomplished, surface loose ends across sessions
Reflect on recent work—capture learnings and surface what to think about next
Post-op check for artifacts, damage, and stale references after agent work
Survey the project and choose what to play next
Create Excalidraw diagrams programmatically. Use when the user wants architecture diagrams, system designs, flowcharts, or any visual diagram with a hand-drawn aesthetic. Outputs editable .excalidraw JSON files. Best for architecture overviews and presentations where the hand-drawn style and drag-and-drop editing add value over Mermaid's text-only approach.
Create diagrams and visualizations using Mermaid.js v11 syntax. Use when generating flowcharts, sequence diagrams, class diagrams, state diagrams, ER diagrams, Gantt charts, user journeys, timelines, or any of 24+ diagram types. Supports CLI rendering to SVG/PNG/PDF, theming, and configuration. Essential for inline documentation diagrams that render natively on GitHub.
| name | vibe-check |
| description | Judgment linter for vibe-coded output — reads the energy of the code, not just correctness. Use when the user says "vibe check", "check this vibe code", "does this hold up", "sanity check this AI code", or after a fast generation session before committing. |
"Vibe check?" — a way of asking what's the energy like right now?
Static linters catch syntax. Code review catches everything but takes time. Vibe-check sits between: a quick judgment pass that asks whether AI-generated code feels like it was written with thought or just generated on autopilot.
Single-agent, one pass. Runs on the diff by default. Surfaces higher-order problems that require reasoning — the things a thoughtful reviewer notices in the first 30 seconds that no regex can detect.
Default: Git diff (staged if present, unstaged otherwise).
# Check current changes
/vibe-check
# Check specific files or directories
/vibe-check src/auth/
Override: Named files or directories passed as arguments.
Each category reads a specific vibe. Findings require AI judgment — if a static linter could catch it, it doesn't belong here.
| Category | The vibe | What it catches |
|---|---|---|
| Earned complexity | "This feels over-engineered" | Abstractions that don't justify themselves, premature generalization, indirection without payoff |
| Prompt-shaped code | "This feels shaped by the prompt, not the problem" | Odd boundaries, misnamed concepts, structure that mirrors how the request was worded rather than the domain |
| Missing skepticism | "This feels like nothing can go wrong" | Happy-path-only logic, unchallenged assumptions, no failure modes considered, silent swallowing of errors |
| Cargo-culted patterns | "This feels ceremonial" | Design patterns applied without justification, boilerplate that adds ceremony without value, framework idioms used out of context |
| Integration blindness | "This feels written in a vacuum" | Duplicates existing utilities, contradicts established codebase patterns, ignores conventions visible in surrounding code |
| Shallow naming | "This feels like nobody thought about what it means" | Names describe implementation (handleData, processItems) instead of intent — a signal the author didn't reason about the domain |
| Security surface | "This feels like it trusts too much" | Auth gaps, unsanitized boundaries, trust assumptions — only the ones requiring judgment, not what semgrep catches |
Each category defaults to a severity level. Users can override per-category.
| Level | Meaning |
|---|---|
| error | Must address before committing |
| warn | Worth considering, not blocking |
| off | Skip this category |
Defaults:
Read the git diff or specified files. Identify the languages and frameworks involved.
For each assertion category (unless severity is off), scan the code and ask the category's question. Apply language-aware judgment:
except, implicit None returns== vs ===, unhandled promise rejections, prototype pollutionunwrap(), clone() to satisfy the borrow checkerGroup findings by category. Each finding includes:
## Prompt-Shaped Code (warn)
- `src/auth/handler.ts:42` — Function boundary splits "validate" and "authorize" because the prompt asked for them separately, but they share all context and should be one operation
- `src/auth/types.ts:8` — `RequestData` is a generic wrapper that mirrors the prompt's language ("request data") rather than the domain concept (`AuthAttempt`)
End with a summary line:
vibe-check: 2 errors, 4 warnings across 3 files
When findings are dense (8+ findings or 4+ categories with hits), suggest escalation:
Dense findings across multiple categories — consider running /review for a deeper pass.
ruff or eslint could catch it, skip it.useEffect cleanup isn't ceremony. Know the difference./review — Full structured review when vibe-check suggests escalation/naming — Deep dive when shallow-naming findings accumulate/testing — When missing-skepticism reveals untested assumptions/sweep — Post-op damage check; vibe-check is pre-commit, sweep is post-mergeskills/FRAMEWORKS.md — Full framework index