DreamCycle

面向 AI agent 的回顾式调试工具：以 Python 命令行扫描 JSON 会话日志，提取 failure/pattern/lesson 信号，识别出现 2 次以上的复发模式及其升降趋势，并为已知失败模式给出带置信度的修复建议；首次运行建立基线，趋势数据保存在本地历史文件中。

qize-auto

热度 29.04

最近更新 2026-05-11 17:47:56

分类：开发依赖：代码执行 / 读写文件应用场景：调试排错 / 数据处理 / 报告撰写 / 工作流

安装技能

复制下面的话术发给你的 Agent，它会通过 secure-skills 检索并安装该技能。

执行命令安装此 skill： npx secure-skills-cli add dreamcycle

Metadata

name: dreamcycle
description: AI Agent Self-Reflection Engine — scan session logs, detect failure patterns, analyze recurrence trends, and suggest automated fixes.
version: 0.1.0
metadata: {"openclaw":{"requires":{"bins":["python3"],"anyBins":["pip3","pip"]},"primaryEnv":"","emoji":"🔄","homepage":"https://github.com/qize-auto/dreamcycle","tags":["agent-self-reflection","debugging","auto-fix","session-analysis","failure-detection"],"category":"development"}}

DreamCycle

Detect, analyze, and auto-repair failure patterns from AI agent session logs.

When to use

Use DreamCycle when:

An agent has been running for a while and you want to check for recurring failures
Session logs are accumulating and you want trend analysis
You suspect the same error keeps happening but no one is tracking it
You want automated fix suggestions for known failure patterns
After making changes to agent configuration, to verify nothing regressed

How it works

DreamCycle is a retrospective debugger for AI agents. It scans your agent's session log files in three phases:

Session Logs → Scan → Analyze → Fix

Scan — Extracts failure, pattern, and lesson signals from JSON session files
Analyze — Detects recurring patterns (appearing 2+ times) and trends (up/down/stable compared to previous scans)
Fix — Maps known failure patterns to automated fix suggestions with confidence scores

Installation

pip install dreamcycle

Usage

# Basic diagnostic: scan last 2 days of sessions
dreamcycle diagnose ~/.openclaw/sessions/

# Scan last 7 days
dreamcycle diagnose ~/my-agent/logs/ --days 7

# JSON output for programmatic use
dreamcycle diagnose ./sessions/ --output json

# Version info
dreamcycle version

Example output

━━━ DreamCycle Diagnose ━━━
Scanned:    42 sessions (last 2 days)
Signals:    156 total (failures: 78, patterns: 52, lessons: 26)
━━━━━━━━━━━━━━━━━━━━━━━━━━

🔄 Top Recurring Patterns
  [12x] "timeout" → ⚡ high
    Fix: add retry_with_backoff (confidence: 85%)
  [8x]  "ModuleNotFoundError" → ⚡ high
    Fix: pip install <module> (confidence: 95%)

📈 Trends
  ↑ timeout: 12x (up from 5x last scan)
  ↓ ModuleNotFoundError: 8x (down from 15x)

🔧 Auto-Fix Summary
  ✅ 2/4 patterns have automated fixes
  ❌ 2 patterns need manual review

Requirements

Only python3 and pip. Zero external dependencies — DreamCycle uses only the Python standard library.

Notes

Session files must be JSON format with a messages array containing role/content fields
First run establishes a baseline; trends appear starting from the second run
Trend data is stored at ~/.dreamcycle/scan_history.json