← 返回 Skills 市场
406
总下载
0
收藏
4
当前安装
1
版本数
在 OpenClaw 中安装
/install feedback-loop
功能描述
Feedback Loop - Collect, analyze, and act on user feedback for continuous agent improvement
使用说明 (SKILL.md)
Feedback Loop Skill
反馈循环器 - 一个完整的反馈收集、分析和改进跟踪系统,用于 OpenClaw 代理的持续优化。
功能概述
核心功能
-
反馈收集(Feedback Collection)
- 显式反馈:用户主动评分、评论
- 隐式反馈:从交互模式自动检测(完成、重试、放弃等)
- 自动检测:基于会话模式智能识别反馈信号
-
反馈分析(Feedback Analysis)
- 聚类分析:按类别、情感、评分分组
- 趋势分析:时间序列趋势检测
- 情感分析:正/负/中性情感分布
- 模式检测:识别 recurring issues、行为模式等
-
改进建议生成(Improvement Suggestions)
- 基于分析结果自动生成可执行的改进建议
- 优先级排序(high/medium/low)
- 包含具体 action items 和预期影响
-
效果跟踪(Effect Tracking)
- 跟踪建议实施进度
- 测量实施前后的影响
- 生成综合报告
安装
# 通过 ClawHub 安装(推荐)
clawhub install feedback-loop
# 或手动安装
cd ~/.openclaw/workspace/skills/feedback-loop
npm install
npm link
使用方法
CLI 命令
1. 提供反馈(provide)
显式反馈:
feedback-loop provide --type explicit --rating 5 --comment "Excellent response!" --category accuracy
隐式反馈:
feedback-loop provide --type implicit --signal completion --sessionId sess123
feedback-loop provide --type implicit --signal retry --metrics '{"retryCount": 3}'
参数说明:
--type: feedback 类型(explicit 或 implicit)--rating: 评分(1-5, thumbs_up, thumbs_down)--comment: 可选评论--category: 反馈类别(accuracy, speed, helpfulness 等)--signal: 隐式信号类型(completion, retry, abandon, correction 等)--sessionId: 会话标识符--metrics: JSON 格式的性能指标--context: JSON 格式的上下文信息
2. 分析反馈(analyze)
# 分析最近一周的反馈
feedback-loop analyze --range week
# 只分析显式反馈
feedback-loop analyze --explicit-only
# 分析最近一个月的数据,输出 JSON
feedback-loop analyze --range month --output json
参数说明:
--range: 时间范围(day, week, month, all)--explicit-only: 仅分析显式反馈--output: 输出格式(json, pretty)
3. 生成建议(suggest)
# 生成最多 5 条建议
feedback-loop suggest --max 5
# 专注于特定类别
feedback-loop suggest --focus quality --max 10
参数说明:
--max: 最大建议数量--focus: 专注的类别--output: 输出格式
4. 跟踪进度(track)
# 跟踪建议实施进度
feedback-loop track fb_123456 --phase implementation --status in_progress
# 标记为已完成
feedback-loop track fb_123456 --phase deployed --status completed --notes "Successfully implemented"
参数说明:
--phase: 实施阶段(planning, implementation, testing, deployed)--status: 状态(in_progress, completed, blocked)--notes: 附加说明--metrics: JSON 格式的进度指标
5. 查看统计(stats)
feedback-loop stats
feedback-loop stats --output json
6. 列出数据(list)
# 列出反馈
feedback-loop list feedback --limit 10
feedback-loop list feedback --type explicit
# 列出建议
feedback-loop list suggestions --status pending
feedback-loop list suggestions --category quality
# 列出跟踪记录
feedback-loop list tracking --phase implementation
7. 生成报告(report)
feedback-loop report
feedback-loop report --output json
8. 导出数据(export)
feedback-loop export --format json --output data.json
feedback-loop export --format csv --output data.csv
编程接口
const FeedbackLoop = require('./src/index');
const fl = new FeedbackLoop();
// 提供反馈
fl.provide({
type: 'explicit',
rating: 5,
comment: 'Great!',
category: 'helpfulness'
});
// 分析
const analysis = fl.analyze({ timeRange: 'week' });
// 生成建议
const suggestions = fl.suggest({ maxSuggestions: 5 });
// 跟踪
fl.track(suggestionId, {
phase: 'implementation',
status: 'in_progress'
});
// 获取统计
const stats = fl.getStats();
// 获取报告
const report = fl.getReport();
触发方式
主动收集
- 在会话结束时自动请求评分
- 定期生成分析报告
- 检测到低满意度时触发改进流程
自动检测
- 高重试率 → 推断用户遇到困难
- 快速完成 → 推断用户满意
- 早期放弃 → 推断响应不符合期望
- 多次追问 → 推断高参与度
数据结构
反馈记录
{
"id": "fb_1234567890_abc123",
"timestamp": "2024-01-15T10:30:00.000Z",
"type": "explicit",
"sessionId": "sess_123",
"rating": 5,
"comment": "Excellent response!",
"category": "accuracy",
"metadata": {},
"source": "cli"
}
隐式反馈
{
"id": "fb_1234567890_def456",
"timestamp": "2024-01-15T10:30:00.000Z",
"type": "implicit",
"sessionId": "sess_123",
"signal": "completion",
"metrics": { "responseTime": 2500 },
"context": { "autoDetected": true },
"inferredSentiment": "positive"
}
改进建议
{
"id": "fb_1234567890_ghi789",
"timestamp": "2024-01-15T10:30:00.000Z",
"title": "Address recurring issues in accuracy",
"description": "5 negative feedback items identified...",
"category": "accuracy",
"priority": "high",
"actionItems": [...],
"expectedImpact": "Reduce negative feedback...",
"status": "pending"
}
最佳实践
- 定期分析:每周运行一次分析,及时发现趋势
- 快速响应:对高优先级建议立即采取行动
- 持续跟踪:记录每个建议的实施进度
- 衡量影响:实施后对比前后数据
- 闭环管理:确保每个反馈都有对应的改进行动
文件结构
feedback-loop/
├── SKILL.md # 技能文档
├── package.json # 项目配置
├── bin/
│ └── cli.js # CLI 入口
├── src/
│ ├── index.js # 主入口
│ ├── storage.js # 数据存储
│ ├── collector.js # 反馈收集
│ ├── analyzer.js # 反馈分析
│ ├── suggester.js # 建议生成
│ └── tracker.js # 效果跟踪
├── data/ # 数据目录(自动生成)
│ ├── feedback.json
│ ├── analysis.json
│ ├── suggestions.json
│ └── tracking.json
└── test/
└── run.js # 测试脚本
注意事项
- 数据存储在
data/目录下,定期备份重要数据 - 建议设置定期清理策略,避免数据文件过大
- 敏感反馈数据应注意隐私保护
版本历史
- 1.0.0 - 初始版本
- 完整的反馈收集功能
- 多维度分析能力
- 智能建议生成
- 效果跟踪系统
安全使用建议
This skill appears internally coherent and implements the feedback pipeline it advertises, but take these precautions before installing:
- Source & provenance: the registry entry lists an unknown owner; review the full repo/files (package.json, bin/cli.js, and all src files) before installing.
- Data sensitivity: it stores all data in data/*.json in the skill folder (unencrypted). If feedback may contain PII or sensitive user content, plan encryption, access controls, or periodic purging.
- Installation: there is no registry-provided installer; follow the README instructions (npm install / npm link) in a controlled environment or sandbox first.
- Audit: run npm audit, run the provided tests, and scan the code for network calls or other changes not present in the supplied files before granting it runtime privileges.
- Autonomy/privacy: the skill can be invoked autonomously by agents (platform default). If you don’t want the agent to record session data automatically, configure agent policies or avoid enabling autonomous invocation for this skill.
If you want more assurance, provide the omitted source files (the truncated portions) or run the CLI in a sandbox and show the runtime behavior for further review.
功能分析
Type: OpenClaw Skill
Name: feedback-loop
Version: 1.0.0
The feedback-loop skill is a standard utility for collecting, analyzing, and tracking agent performance metrics. The code uses local JSON files for persistence within its own directory and lacks any network dependencies, obfuscation, or dangerous execution sinks like eval or child_process.exec. Analysis of SKILL.md and the source code (src/storage.js, src/collector.js) shows no evidence of malicious intent, data exfiltration, or prompt injection attacks.
能力评估
Purpose & Capability
Name/description align with the included modules (collector, analyzer, suggester, tracker, storage). There are no extra credentials, binaries, or unrelated capabilities requested that would be unexpected for a feedback/analytics skill.
Instruction Scope
SKILL.md and the CLI instruct only feedback-related operations (provide, analyze, suggest, track, export). The code reads/writes only local files under the skill's data/ directory and does not reference external network endpoints, unrelated system paths, or undeclared environment variables.
Install Mechanism
Registry metadata states 'no install spec' (instruction-only) but the package contains package.json and a CLI and recommends npm install/npm link. No installer is provided by the registry entry — installation requires running npm commands locally. This is explainable but worth noting: the skill will write code and data into the local skill workspace when installed.
Credentials
The skill requires no environment variables or credentials. All storage is local to the skill (data/*.json). There are no requests for unrelated secrets or access to other skills' config.
Persistence & Privilege
The skill persists feedback, analyses, suggestions, and tracking to a local data/ folder under the skill. 'always' is false and autonomous invocation is allowed (platform default). Persistent local storage of potentially sensitive feedback is appropriate for the skill's purpose but should be considered by users for privacy and retention policies.
如何使用
- 确保已安装 OpenClaw(本地或 Docker 部署)
- 在对话框中输入安装命令:
/install feedback-loop - 安装完成后,直接呼叫该 Skill 的名称或使用
/feedback-loop触发 - 根据 Skill 的参数说明提供必要输入,即可获得结构化输出
版本历史
v1.0.0
Initial release of Feedback Loop skill.
- Collect explicit and implicit user feedback with CLI commands and programming interface.
- Analyze feedback with clustering, trend, sentiment, and pattern detection.
- Automatically generate prioritized improvement suggestions with actionable items.
- Track progress and effects of implemented suggestions.
- Generate statistics and reports; export data in JSON/CSV.
- Organize all feedback, suggestions, and tracking data for ongoing agent optimization.
元数据
常见问题
Feedback Loop 是什么?
Feedback Loop - Collect, analyze, and act on user feedback for continuous agent improvement. 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 406 次。
如何安装 Feedback Loop?
在 OpenClaw 或 Claude Code 对话框中运行命令「/install feedback-loop」即可一键安装,无需额外配置。
Feedback Loop 是免费的吗?
是的,Feedback Loop 完全免费,采用 MIT-0 许可证,可自由下载、安装和使用。
Feedback Loop 支持哪些平台?
Feedback Loop 跨平台运行,可在任意部署了 OpenClaw / Claude Code 的环境中使用(cross-platform)。
谁开发了 Feedback Loop?
由 haidong(@harrylabsj)开发并维护,当前版本 v1.0.0。
推荐 Skills