refresh

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Vibe Check Refresh

<objective> Run a complete fresh assessment and compare against the previous results. This is NOT a shallow re-check — it's a full re-analysis that catches new issues, validates fixes, and removes false positives. </objective> <architecture>

You are the orchestrator. Same as

/check

, but you also track changes.

/refresh (orchestrator)
    │
    ├── Phase 1: Load Previous State
    │   └── Read metadata.json, note old scores and items
    │
    ├── Phase 2: Clear Analysis (fresh start)
    │   └── Delete analysis/ folder contents
    │
    ├── Phase 3: Full Re-mapping
    │   └── Spawn: mapper agent (runs npm audit, etc.)
    │       └── Writes: .vibe-check/analysis/*.md
    │
    ├── Phase 4: Full Re-assessment
    │   └── Spawn: assessor agents (parallel)
    │       └── Each writes NEW checklist items
    │
    ├── Phase 5: Compare & Reconcile
    │   └── Compare new items vs old items
    │   └── DELETE items that were false positives
    │   └── Track improvements and regressions
    │
    ├── Phase 6: Update Files
    │   └── Rewrite all output files with new findings
    │
    └── Phase 7: Show Progress
        └── Display what changed

Key difference from check: You compare new results against old, track deltas, and explicitly remove items that shouldn't exist.

</architecture>

<objective> 运行完整的全新评估并与之前的结果进行对比。这不是浅层的重新检查——而是全面的重新分析，能够发现新问题、验证修复效果并移除误报项。 </objective> <architecture>

你是编排器。与

/check

功能相同，但你还需要跟踪变更情况。

/refresh (orchestrator)
    │
    ├── Phase 1: Load Previous State
    │   └── 读取metadata.json，记录旧的得分和评估项
    │
    ├── Phase 2: Clear Analysis (fresh start)
    │   └── 删除analysis/文件夹中的内容
    │
    ├── Phase 3: Full Re-mapping
    │   └── 启动：mapper agent（运行npm audit等检查）
    │       └── 写入：.vibe-check/analysis/*.md
    │
    ├── Phase 4: Full Re-assessment
    │   └── 启动：assessor agents（并行执行）
    │       └── 每个agent写入新的检查清单项
    │
    ├── Phase 5: Compare & Reconcile
    │   └── 对比新评估项与旧评估项
    │   └── 删除属于误报的项
    │   └── 跟踪改进和退化情况
    │
    ├── Phase 6: Update Files
    │   └── 用新的评估结果重写所有输出文件
    │
    └── Phase 7: Show Progress
        └── 展示变更内容

与/check的核心区别：你会将新结果与旧结果进行对比，跟踪差异，并明确移除不应存在的评估项。

</architecture>

Reference Files

参考文件

Read these as needed:

```
agents/mapper.md
```
— Spawn instructions for codebase mapper
```
agents/assessor.md
```
— Spawn instructions for domain assessors

For domain criteria, read

../check/reference/domains.md

if available in a sibling directory. If not found, fall back to the inline domain summary below.

<domain_summary>

Security (max 25): Secrets management, authentication, input validation, dependency security, HTTPS.

Discoverability (max 20): Meta tags, OpenGraph, Twitter cards, sitemap, robots.txt, semantic HTML.

Analytics (max 15): Visitor tracking, error tracking, conversion tracking.

Platform (max 15): Hosting compatibility, complexity check, cost signals, managed services.

Reliability (max 15): Backups, error handling, database connections, health checks.

Legal (max 10): Privacy policy, terms of service, cookie consent, user data deletion.

AI Security (max 20, conditional): Prompt injection, function calling safety, WebSocket origin, plugin security, context isolation.

</domain_summary>

按需阅读以下文件：

```
agents/mapper.md
```
— 代码库映射器的启动说明
```
agents/assessor.md
```
— 领域评估器的启动说明

关于领域评估标准，如果同级目录下存在

../check/reference/domains.md

则读取该文件；若未找到，则使用下方的内联领域摘要。

<domain_summary>

安全（满分25）：密钥管理、身份验证、输入验证、依赖安全性、HTTPS。

可发现性（满分20）：元标签、OpenGraph、Twitter卡片、站点地图、robots.txt、语义化HTML。

分析能力（满分15）：访客跟踪、错误跟踪、转化跟踪。

平台适配（满分15）：托管兼容性、复杂度检查、成本信号、托管服务。

可靠性（满分15）：备份、错误处理、数据库连接、健康检查。

合规性（满分10）：隐私政策、服务条款、Cookie同意、用户数据删除。

AI安全（满分20，可选）：提示注入防护、函数调用安全、WebSocket源验证、插件安全、上下文隔离。

</domain_summary>

Prerequisites

前置条件

Requires existing

.vibe-check/metadata.json

. If not found:

No existing assessment found.
Run /check first to create an initial assessment.

需要已存在

.vibe-check/metadata.json

文件。若未找到：

未找到现有评估记录。
请先运行/check创建初始评估。

Process

流程

Phase 1: Load Previous State

阶段1：加载历史状态

Read

.vibe-check/metadata.json

and note:

Previous analysis date
Previous score
Previous item list (IDs, slugs, statuses)
Deployment and compliance context (reuse these — don't re-ask)

Store this for comparison later.

读取

.vibe-check/metadata.json

并记录：

上次分析日期
上次评估得分
上次评估项列表（ID、别名、状态）
部署与合规上下文（复用这些信息——无需重新询问）

将这些信息存储起来以便后续对比。

Phase 2: Clear Analysis Folder

阶段2：清空分析文件夹

Delete contents of

.vibe-check/analysis/

to ensure fresh data:

bash

rm -rf .vibe-check/analysis/*

Do NOT delete checklist items yet — you need them for comparison.

删除

.vibe-check/analysis/

目录下的内容以确保数据是全新的：

bash

rm -rf .vibe-check/analysis/*

请勿立即删除检查清单项——你需要这些项用于对比。

Phase 3: Full Re-mapping

阶段3：完整重新映射

If subagent spawning is available, spawn the mapper agent:

Task: Map codebase for readiness assessment

Read agents/mapper.md for your instructions.

Explore this codebase and write analysis files to .vibe-check/analysis/

This is a REFRESH run. Be thorough — run all checks including npm audit.

Return confirmation only when complete.

If subagent spawning is NOT available, perform the mapping yourself:

Read
```
agents/mapper.md
```
for the exploration process
Follow compact mode or standard mode steps depending on codebase size
Write all 13 analysis files to
```
.vibe-check/analysis/
```
Continue to Phase 4

Wait for confirmation. The mapper runs the full analysis including:

```
npm audit
```
/
```
yarn audit
```
/
```
pnpm audit
```
Git status checks for .env files
All other exploration steps

如果支持子Agent启动，启动mapper agent：

任务：为就绪性评估映射代码库

请阅读agents/mapper.md获取你的执行说明。

探索此代码库并将分析文件写入.vibe-check/analysis/

这是一次REFRESH运行。请全面检查——包括运行npm audit等所有检查项。

完成后仅返回确认信息。

如果不支持子Agent启动，则自行执行映射操作：

阅读
```
agents/mapper.md
```
了解探索流程
根据代码库大小选择紧凑模式或标准模式执行步骤
将所有13个分析文件写入
```
.vibe-check/analysis/
```
进入阶段4

等待确认信息。映射器会执行完整分析，包括：

```
npm audit
```
/
```
yarn audit
```
/
```
pnpm audit
```
检查.env文件的Git状态
所有其他探索步骤

Phase 4: Full Re-assessment

阶段4：完整重新评估

Spawn

assessor

agents for each domain — exactly like

/check

If subagent spawning is available, spawn for each domain using the same prompts as the check skill. Use item numbers starting at 100+ for new items to avoid conflicts during comparison.

Example for Security Assessor:

Task: Assess security domain

Read agents/assessor.md for your instructions.
Your domain assignment is: security

Load analysis files from .vibe-check/analysis/
Read agents/assessor.md for criteria.

This is a REFRESH run. Evaluate fresh — ignore previous checklist items.
Write NEW items to .vibe-check/checklist/ with fresh item numbers starting at 100.

Return score summary only.

If subagent spawning is NOT available, assess each domain yourself using the criteria in

agents/assessor.md

Spawn all domain assessors (security, discoverability, analytics, platform, reliability, legal). Spawn AI security assessor only if mapper indicated AI patterns detected.

为每个领域启动

assessor

agent——与

/check

流程完全一致：

如果支持子Agent启动，使用与check技能相同的提示为每个领域启动agent。新评估项的编号从100+开始，以避免对比时产生冲突。

安全评估器示例：

任务：评估安全领域

请阅读agents/assessor.md获取你的执行说明。
你的领域分配是：security

从.vibe-check/analysis/加载分析文件
阅读agents/assessor.md获取评估标准。

这是一次REFRESH运行。请基于全新数据评估——忽略之前的检查清单项。
将新的评估项写入.vibe-check/checklist/，编号从100开始。

仅返回得分摘要。

如果不支持子Agent启动，则使用

agents/assessor.md

中的标准自行评估每个领域。

为所有领域启动评估器（安全、可发现性、分析能力、平台适配、可靠性、合规性）。仅当映射器检测到AI模式时，才启动AI安全评估器。

Phase 5: Compare & Reconcile

阶段5：对比与协调

Now you have:

Old items:

.vibe-check/checklist/item-001-*.md

through

item-0XX-*.md

New items:

.vibe-check/checklist/item-100-*.md

through

item-1XX-*.md

For each OLD item, check if a corresponding NEW item exists (match by slug/topic):

Old Status	New Status	Action
Fail	Pass (or no new item)	Improved — delete old item
Fail	Fail	Unchanged — update old item with new findings
Fail	N/A	Now N/A — delete old item (capability removed or context changed)
Pass	Fail	Regressed — update old item
Pass	N/A	Now N/A — note in delta (not a regression)
Unknown	Pass/Fail	Resolved — update status
Unknown	N/A	Now N/A — delete old item
N/A	Fail	Now applicable — treat as new issue
N/A	Pass	Now applicable — no action needed
Any	No match	False positive or N/A — DELETE the old item

For NEW items with no OLD match:

New issue discovered — keep with renumbered ID

After reconciliation:

Delete all
```
item-100+
```
temporary files
Renumber remaining items sequentially
Update
```
checklist/index.md
```

现在你拥有：

旧评估项：

.vibe-check/checklist/item-001-*.md

至

item-0XX-*.md

新评估项：

.vibe-check/checklist/item-100-*.md

至

item-1XX-*.md

针对每个旧评估项，检查是否存在对应的新评估项（通过别名/主题匹配）：

旧状态	新状态	操作
失败	通过（或无对应新项）	已改进 — 删除旧评估项
失败	失败	未变更 — 用新发现更新旧评估项
失败	不适用	变为不适用 — 删除旧评估项（功能已移除或上下文变更）
通过	失败	已退化 — 更新旧评估项
通过	不适用	变为不适用 — 在差异报告中注明（不属于退化）
未知	通过/失败	已解决 — 更新状态
未知	不适用	变为不适用 — 删除旧评估项
不适用	失败	变为适用 — 视为新问题
不适用	通过	变为适用 — 无需操作
任意	无匹配项	误报或不适用 — 删除旧评估项

对于无匹配旧项的新评估项：

发现新问题 — 保留并重新编号

协调完成后：

删除所有
```
item-100+
```
临时文件
将剩余评估项按顺序重新编号
更新
```
checklist/index.md
```

Phase 6: Update Files

阶段6：更新文件

Rewrite all output files:

metadata.json — New scores with
```
previousAnalysis
```
field:

json

{
  "analysisDate": "{today}",
  "score": {new score},
  "previousAnalysis": {
    "date": "{old date}",
    "score": {old score}
  },
  ...
}

summary.md — Updated with progress section
report.md — Full fresh report
action-plan.md — Reprioritized based on current state
checklist/index.md — Updated item list

重写所有输出文件：

metadata.json — 新增
```
previousAnalysis
```
字段的新得分：

json

{
  "analysisDate": "{today}",
  "score": {new score},
  "previousAnalysis": {
    "date": "{old date}",
    "score": {old score}
  },
  ...
}

summary.md — 更新进度部分
report.md — 生成全新的完整报告
action-plan.md — 根据当前状态重新排序优先级
checklist/index.md — 更新评估项列表

Phase 7: Terminal Output & Discussion

阶段7：终端输出与讨论

Use consistent visual patterns for output. Display progress with comparison bars:

┌──────────────────────────────────────────────┐
│                                              │
│   VIBE CHECK REFRESH                         │
│                                              │
│   Score: {new_score}/100                     │
│   {progress_bar}  {band}                     │
│                                              │
└──────────────────────────────────────────────┘

PROGRESS
═══════════════════════════════════════════════

Previous:  {old_score}/100  {old_bar}  {old_band}
Current:   {new_score}/100  {new_bar}  {new_band}
                            ─────────────────────
Change:                     {+/-delta} points  {▲/▼}

SUMMARY
─────────────────────────────────────────────────
  ✓ {N} items fixed
  ○ {N} items unchanged
  ✗ {N} items regressed
  + {N} new issues found
  − {N} false positives removed
  ○ {N} items now N/A

{If fixed items:}
IMPROVEMENTS
─────────────────────────────────────────────────
✓  {item title}
✓  {item title}

{If regressions:}
REGRESSIONS
─────────────────────────────────────────────────
✗  {item title} — was passing, now failing

{If still failing:}
STILL FAILING
─────────────────────────────────────────────────
●  {title} — {one-line description}
●  {title} — {one-line description}

┌─ NEXT ──────────────────────────────────────┐
│                                             │
│  • Review: .vibe-check/summary.md           │
│  • Fix:    /fix                             │
│  • Discuss: /discuss                        │
│                                             │
└─────────────────────────────────────────────┘

Then offer discussion (same as

/check

Would you like to:
1. Discuss the changes — Ask about specific items
2. Start fixing — Pick an item to work on
3. Done for now

使用统一的视觉格式输出。用对比进度条展示改进情况：

┌──────────────────────────────────────────────┐
│                                              │
│   VIBE CHECK REFRESH                         │
│                                              │
│   得分: {new_score}/100                     │
│   {progress_bar}  {band}                     │
│                                              │
└──────────────────────────────────────────────┘

进度情况
═══════════════════════════════════════════════

上次评估:  {old_score}/100  {old_bar}  {old_band}
当前评估:   {new_score}/100  {new_bar}  {new_band}
                            ─────────────────────
变化:                     {+/-delta} 分  {▲/▼}

摘要
─────────────────────────────────────────────────
  ✓ {N} 项已修复
  ○ {N} 项无变化
  ✗ {N} 项出现退化
  + {N} 发现新问题
  − {N} 移除误报项
  ○ {N} 项变为不适用

{如果有已修复项:}
改进项
─────────────────────────────────────────────────
✓  {评估项标题}
✓  {评估项标题}

{如果有退化项:}
退化项
─────────────────────────────────────────────────
✗  {评估项标题} — 之前通过，现在失败

{如果仍有失败项:}
仍未通过项
─────────────────────────────────────────────────
●  {标题} — {单行描述}
●  {标题} — {单行描述}

┌─ 下一步 ──────────────────────────────────────┐
│                                             │
│  • 查看: .vibe-check/summary.md           │
│  • 修复:    /fix                             │
│  • 讨论: /discuss                        │
│                                             │
└─────────────────────────────────────────────┘

然后提供与

/check

相同的讨论选项：

你想要：
1. 讨论变更情况 — 询问具体评估项
2. 开始修复 — 选择一个评估项进行处理
3. 暂时结束

Critical Rules

关键规则

RUN THE FULL PIPELINE. Spawn mapper and assessors. Don't try to shortcut by just re-reading old analysis files.

DELETE FALSE POSITIVES. If the new assessment doesn't flag something, remove it. Don't keep stale items around.

FRESH ANALYSIS FILES. Clear and regenerate analysis/ folder. Old analysis data causes stale findings.

RENUMBER ITEMS. After reconciliation, items should be numbered 001, 002, 003... sequentially.

TRACK DELTAS. The whole point is showing progress. Make sure to report what changed.

执行完整流程。启动mapper和assessor agent。不要试图通过仅读取旧分析文件来走捷径。

删除误报项。如果新评估未标记某问题，则将其移除。不要保留过时的评估项。

生成全新分析文件。清空并重新生成analysis/文件夹。旧的分析数据会导致过时的评估结果。

重新编号评估项。协调完成后，评估项应按001、002、003...的顺序连续编号。

跟踪差异。此功能的核心是展示进度。确保准确报告所有变更内容。

Context

上下文