克劳德策展人
- 作者仓库星标 0
- 作者仓库 skills-registry
- 领域
- 安全
- 作者声明的智能体
- 未找到明确声明;不据此推断已兼容或已测试
- 静态检查
- 88 / 100 · 启发式扫描,不代表运行安全
- 运行环境证据
- 未执行;当前没有隔离运行测试证据
- 作者 / 版本 / 许可证
- @tomevault-io · 未声明许可证
- 上下文用量评级
- 低消耗
- 接入复杂程度
- 需简单配置
- 是否需要外部接口密钥
- 不需要
- 兼容的系统
- macOS · Linux · Windows
- 运行环境要求
- 未检出明确要求
- 检测到的文件与系统行为
-
- 只读
- 允许写入 / 修改
- 命令执行
- 读取环境变量
- 网络行为
- 仅限本地
- 安装命令数
- 无(仅作为资料)
需要注意: 未声明工具权限范围;实际权限由运行环境决定。
查找存储库中的所有项目内存工件和已知的 Claude 用户内存位置。当用户明确要求多存储库覆盖(例如,“检查所有工作区”、“审核工作区内存”)时,请将发现扩展到 WORKSPACE_DIRS 环境变量中配置的工作区根或使用默认示例。发现运行在所有匹配的工件上,无需采样或截断。该技能处理完整列表以满足 claudemd-curator 合同。
对于每个发现的项目内存工件(`CLAUDE.md`、`AGENTS.md`、`.claude.local.md`、`.claude/rules/*.md`),根据质量标准进行评估。请参阅 References/quality-criteria.md 了解详细的标准。在最终评分之前,将`~/.claude/CLAUDE.md`(如果存在)与每个项目级别的`CLAUDE.md`进行比较。将重复的命令、重叠的现行规则和相互冲突的指令暴露为潜在的偏差。报告必须命名全局文件和每个受影响的项目路径;不要默默地合并或重写全局内容。
当存储库包含市场/插件元数据时,针对现有市场真实信号运行插件感知审计,而不是产生重复的漂移事实。 使用 `marketplace-tour` 的 `capability-snapshot` 输出(如果可用),并将发现的内存伪影与 `METADATA_DRIFT`、`CONTENT_DRIFT` 和 `STALE_<N>d` 信号交叉引用。
在进行任何更新之前始终输出质量报告。格式:验证:报告包括文件计数、平均分数、每个发现的工件、具体问题和建议的添加内容。失败条件:如果建议未指定目标文件和原因,则不要继续更新。
输出质量报告后,更新前请用户确认。更新指南(关键):仅提出有针对性的添加 - 关注真正有用的信息:分析过程中发现的命令或工作流程代码中发现的陷阱或不明显的模式
用户批准后,使用编辑工具应用更改。保留现有的内容结构。验证:重新阅读每个编辑过的工件,并确认批准的文本没有进行不相关的重写。失败条件:如果批准缺失或不明确,请在编辑之前停止并要求更新特定目标文件。
## MANDATORY ACTIVATION
Use this skill when the user asks for:
- "audit CLAUDE.md" or "check CLAUDE.md"
- "improve project memory" or "optimize memory"
- "sync AGENTS.md" or "Codex AGENTS.md sync"
- "update .claude/rules" or "review .claude/rules"
- "memory quality report"
## FAILURE CONDITIONS
**Skipping the claudemd-curator skill entirely** leaves project memory in a degraded state, causing:
- **Degraded memory quality**: Stale, incomplete, or conflicting instructions in `CLAUDE.md`, `AGENTS.md`, `.claude.local.md`, and `.claude/rules/*.md`
- **Missed rule updates**: Critical commands, patterns, or architecture changes remain undocumented
- **Unsynced cross-tool memory**: Claude Code and Codex/Codeium agents diverge when `CLAUDE.md` and `AGENTS.md` fall out of sync
- **Audit omissions**: Marketplace drift signals (from `marketplace-tour`) go unconsumed, leaving memory guidance mismatched to current plugin capabilities
- **Downstream agent failures**: Other agents operate on outdated commands, obsolete architecture notes, or missing gotchas
**Severity & Escalation**: Treat skipped memory curation as **high severity** when the task depends on repository instructions, cross-tool consistency, or accurate project context. Escalate when memory staleness blocks task completion or causes repeated agent failures.
**Per-Phase Failure Details**: See individual phase sections below for specific failure conditions during discovery (Phase 1, line 52), quality assessment (Phase 2, line 98), plugin mode (Phase 2.5, line 108), report output (Phase 3, line 159), targeted updates (Phase 4, line 204), and applying updates (Phase 5, line 212). These describe granular failure modes; this section summarizes the global impact when claudemd-curator is bypassed entirely.
# claudemd-curator
Audit, evaluate, and improve project-memory artifacts across a codebase so Claude Code (and Codex / Codeium / ChatGPT, which read `AGENTS.md`) have optimal project context.
**This skill can write to memory files.** After presenting a quality report and getting user approval, it updates `CLAUDE.md`, `AGENTS.md`, `.claude.local.md`, or files under `.claude/rules/` with targeted improvements.
`MEMORY.md` (the auto-memory index under `~/.claude/projects/*/memory/`) is recognized but **never rewritten by this skill** — it is owned by the Claude Code memory subsystem and has its own format with frontmatter.
## Workflow
### Phase 1: Discovery
Find all project-memory artifacts in the repository and known Claude user-memory locations. When the user explicitly asks for multi-repo coverage (e.g., "check all workspaces", "audit workspace memory"), extend discovery to workspace roots configured in WORKSPACE_DIRS environment variable or use default examples.
```bash
{
find . \( -name "CLAUDE.md" -o -name "AGENTS.md" -o -name ".claude.md" -o -name ".claude.local.md" \) 2>/dev/null
find . -path "*/.claude/rules/*.md" 2>/dev/null
# Workspace mode: only when user explicitly requests multi-repo coverage
# Configure via WORKSPACE_DIRS environment variable, e.g.:
# export WORKSPACE_DIRS="$HOME/framework $HOME/qyl $HOME/marketplaces"
# Or update these example paths to match your actual workspace locations.
if [ "${WORKSPACE_MODE:-false}" = "true" ]; then
workspace_roots="${WORKSPACE_DIRS:-$HOME/framework $HOME/qyl $HOME/marketplaces}"
for workspace_root in $workspace_roots; do
[ -d "$workspace_root" ] || continue
echo "[workspace-scan] $workspace_root" >&2
find "$workspace_root" -maxdepth 5 \( -name "CLAUDE.md" -o -name "AGENTS.md" -o -name ".claude.md" -o -name ".claude.local.md" \) 2>/dev/null
find "$workspace_root" -maxdepth 6 -path "*/.claude/rules/*.md" 2>/dev/null
done
fi
find ~/.claude -maxdepth 1 -name "CLAUDE.md" 2>/dev/null
find ~/.claude/projects -path "*/memory/MEMORY.md" 2>/dev/null
}
```
Discovery runs across all matching artifacts with no sampling or truncation. The skill processes the complete list to satisfy the claudemd-curator contract.
**Workspace Mode Activation:** Set `WORKSPACE_MODE=true` when the user asks for multi-repo coverage. The report must include a workspace discovery section listing:
- Which workspace roots from WORKSPACE_DIRS (or the default list) were checked
- Which roots existed on disk
- How many artifacts each contributing workspace yielded
- Total artifacts across all workspaces
**Verification:** At least one project-memory artifact is discovered, and every advertised location type that exists on disk appears in the candidate list.
**Failure condition:** If no artifacts are found, stop and report the repository path, workspace roots checked (if workspace mode was active), and discovery commands instead of inventing memory updates.
**File Types & Locations:**
| Type | Location | Purpose |
|------|----------|---------|
| Project root (Claude) | `./CLAUDE.md` | Primary project context for Claude Code (checked into git, shared with team) |
| Project root (Codex) | `./AGENTS.md` | Project context for Codex / OpenAI / Codeium tooling — peer of CLAUDE.md, treated as authoritative for those tools |
| Local overrides | `./.claude.local.md` | Personal/local settings (gitignored, not shared) |
| Auto-loaded rules | `./.claude/rules/*.md` | Every `.md` under this directory is auto-loaded by Claude Code into the system prompt |
| Global defaults | `~/.claude/CLAUDE.md` | User-wide defaults across all projects |
| Package-specific | `./packages/*/CLAUDE.md` | Module-level context in monorepos |
| Subdirectory | Any nested location | Feature/domain-specific context |
| Auto-memory index | `~/.claude/projects/*/memory/MEMORY.md` | Auto-memory pointer file — **read-only awareness**, this skill never rewrites it |
**Notes:**
- Claude auto-discovers `CLAUDE.md` files in parent directories, making monorepo setups work automatically.
- `AGENTS.md` is the convention used by Codex, OpenAI Codex CLI, and Codeium / Continue. When both `CLAUDE.md` and `AGENTS.md` exist, treat them as a pair: the same facts may need to live in both, but each file should be self-contained because each tool reads only its own.
- `.claude/rules/*.md` files are loaded as a flat set; ordering across files is not guaranteed. Keep each rule file independently coherent.
### Phase 2: Quality Assessment
For each discovered project-memory artifact (`CLAUDE.md`, `AGENTS.md`, `.claude.local.md`, `.claude/rules/*.md`), evaluate against quality criteria. See [references/quality-criteria.md](references/quality-criteria.md) for detailed rubrics.
Before final scoring, compare `~/.claude/CLAUDE.md` (when present) with every project-level `CLAUDE.md`. Surface duplicated commands, overlapping standing rules, and conflicting instructions as potential drift. The report must name the global file and each affected project path; do not silently merge or rewrite global content.
**Quick Assessment Checklist:**
| Criterion | Weight | Check |
|-----------|--------|-------|
| Commands/workflows documented | High | Are build/test/deploy commands present? |
| Architecture clarity | High | Can Claude understand the codebase structure? |
| Non-obvious patterns | Medium | Are gotchas and quirks documented? |
| Conciseness | Medium | No verbose explanations or obvious info? |
| Currency | High | Does it reflect current codebase state? |
| Actionability | High | Are instructions executable, not vague? |
**Quality Scores:**
- **A (90-100)**: Comprehensive, current, actionable
- **B (70-89)**: Good coverage, minor gaps
- **C (50-69)**: Basic info, missing key sections
- **D (30-49)**: Sparse or outdated
- **F (0-29)**: Missing or severely outdated
**Verification:** Every discovered writable artifact has a score and notes for each checklist criterion, and the report includes a global/project overlap note when both global and project `CLAUDE.md` files exist.
**Failure condition:** If an artifact cannot be read or scored, mark it blocked with the exact path and continue scoring the remaining artifacts.
### Phase 2.5: Plugin Mode
When the repository includes marketplace/plugin metadata, run plugin-aware audit against existing marketplace truth signals instead of producing duplicate drift facts. Consume `capability-snapshot` output from `marketplace-tour` when available and cross-reference discovered memory artifacts with `METADATA_DRIFT`, `CONTENT_DRIFT`, and `STALE_<N>d` signals.
Use Phase 2.5 to explain whether memory guidance matches the current plugin registry, manifests, command names, skill names, and documented capability surface. The skill must not re-emit duplicate `METADATA_DRIFT`, `CONTENT_DRIFT`, or `STALE_<N>d` signals; it should link or cite the existing signal and describe the memory-file impact.
**Verification:** Plugin-mode output lists the capability-snapshot source, every consumed marketplace-tour signal, and the memory artifacts affected by each signal.
**Failure condition:** If capability-snapshot data is unavailable, mark Phase 2.5 as skipped with the exact missing source and continue the normal memory-quality report.
### Phase 3: Quality Report Output
**ALWAYS output the quality report BEFORE making any updates.**
Format:
```
## Project Memory Quality Report
### Summary
- Files found: X (CLAUDE.md, AGENTS.md, .claude.local.md, .claude/rules/*.md discovered)
- Average score: X/100
- Files needing update: X
### File-by-File Assessment
#### 1. ./CLAUDE.md (Project Root - Claude)
**Score: XX/100 (Grade: X)**
| Criterion | Score | Notes |
|-----------|-------|-------|
| Commands/workflows | X/20 | ... |
| Architecture clarity | X/20 | ... |
| Non-obvious patterns | X/15 | ... |
| Conciseness | X/15 | ... |
| Currency | X/15 | ... |
| Actionability | X/15 | ... |
**Issues:**
- [List specific problems]
**Recommended additions:**
- [List what should be added]
#### 2. ./AGENTS.md (Project Root - Codex/OpenAI/Codeium)
[Same format, scored against the same rubric]
#### 3. ./.claude.local.md (Local overrides)
[Same format if present]
#### 4. ./.claude/rules/<rule>.md (Auto-loaded rules)
[One assessment per discovered rule file]
[Repeat for all discovered memory artifacts]
...
```
**Verification:** The report includes file count, average score, every discovered artifact, concrete issues, and recommended additions.
**Failure condition:** If recommendations do not name the target file and reason, do not proceed to updates.
### Phase 4: Targeted Updates
After outputting the quality report, ask user for confirmation before updating.
**Update Guidelines (Critical):**
1. **Propose targeted additions only** - Focus on genuinely useful info:
- Commands or workflows discovered during analysis
- Gotchas or non-obvious patterns found in code
- Package relationships that weren't clear
- Testing approaches that work
- Configuration quirks
2. **Keep it minimal** - Avoid:
- Restating what's obvious from the code
- Generic best practices already covered
- One-off fixes unlikely to recur
- Verbose explanations when a one-liner suffices
3. **Show diffs** - For each change, show:
- Which project-memory artifact to update
- The specific addition (as a diff or quoted block)
- Brief explanation of why this helps future sessions
**Diff Format:**
```markdown
### Update: ./CLAUDE.md
**Why:** Build command was missing, causing confusion about how to run the project.
```diff
+ ## Quick Start
+
+ ```bash
+ npm install
+ npm run dev # Start development server on port 3000
+ ```
```
```
**Verification:** Each proposed update is a minimal diff with a target file and a one-line reason tied to the quality report.
**Failure condition:** If a proposed change is generic, stale, or not backed by repo evidence, remove it from the proposal.
### Phase 5: Apply Updates
After user approval, apply changes using the Edit tool. Preserve existing content structure.
**Verification:** Re-read each edited artifact and confirm the approved text landed without unrelated rewrites.
**Failure condition:** If approval is missing or ambiguous, stop before editing and ask for the specific target files to update.
## Templates
See [references/templates.md](references/templates.md) for CLAUDE.md templates by project type.
## Common Issues to Flag
1. **Stale commands**: Build commands that no longer work
2. **Missing dependencies**: Required tools not mentioned
3. **Outdated architecture**: File structure that's changed
4. **Missing environment setup**: Required env vars or config
5. **Broken test commands**: Test scripts that have changed
6. **Undocumented gotchas**: Non-obvious patterns not captured
## User Tips to Share
When presenting recommendations, remind users:
- **`#` key shortcut**: During a Claude session, press `#` to have Claude auto-incorporate learnings into CLAUDE.md
- **Keep it concise**: CLAUDE.md should be human-readable; dense is better than verbose
- **Actionable commands**: All documented commands should be copy-paste ready
- **Use `.claude.local.md`**: For personal preferences not shared with team (add to `.gitignore`)
- **Global defaults**: Put user-wide preferences in `~/.claude/CLAUDE.md`
## What Makes a Great CLAUDE.md
**Key principles:**
- Concise and human-readable
- Actionable commands that can be copy-pasted
- Project-specific patterns, not generic advice
- Non-obvious gotchas and warnings
**Recommended sections** (use only what's relevant):
- Commands (build, test, dev, lint)
- Architecture (directory structure)
- Key Files (entry points, config)
- Code Style (project conventions)
- Environment (required vars, setup)
- Testing (commands, patterns)
- Gotchas (quirks, common mistakes)
- Workflow (when to do what)
---
> Source: [ANcpLua/ancplua-claude-plugins](https://github.com/ANcpLua/ancplua-claude-plugins) — distributed by [TomeVault](https://tomevault.io).
<!-- tomevault:4.0:skill_md:2026-05-23 -->
按说明文件中的顺序排列。
强制激活
失效情况
第一阶段:发现
第二阶段:质量评估
阶段 2.5:插件模式
文件/命令 -> CLAUDE.md · AGENTS.md · .claude.local.md · .claude/rules/*.md · marketplace-tour · .claude/rules/ · MEMORY.md · ~/.claude/projects/*/memory/
内容校验值 -> 0316106596d2
当前原文的自动摘译,尚未人工审校。
强制激活
当用户请求以下内容时使用此技能:“审核 CLAUDE.md”或“检查 CLAUDE.md”“改进项目内存”或“优化内存”“同步 AGENTS.md”或“Codex AGENTS.md 同步”“更新 .claude/rules”或“审查 .claude/rules”
失效情况
完全跳过 claudemd-curator 技能会使项目内存处于降级状态,从而导致: 内存质量下降:
CLAUDE.md、AGENTS.md、.claude.local.md和.claude/rules/*.md中的指令陈旧、不完整或冲突 错过规则更新:关键命令、模式或架构更改仍未记录 未同步的跨工具内存:Claude 代码和当CLAUDE.md和AGENTS.md不同步时,Codex/Codeium 代理会出现分歧