AI Agent 引导系统
AI Agent Guidance System
为 AI 编码 Agent 生成结构化行动指引，驱动自主代码改善循环
子问题
1.Phase detection and milestone tracking
2.Strategy computation for optimal fix ordering
3.Skill document generation for multiple AI IDEs
4.Risk flag and reminder system
5.Fixer leverage estimation for auto-fix coverage
6.Reminder decay to prevent Agent information overload
7.Headline generation adapting to project lifecycle phase
8.Wontfix debt trend analysis and stale decision detection
各项目的解法1 solutions
横向对比
| 维度 | Desloppify |
|---|---|
| 引导架构 | Narrative 纯函数管道：state → phase/actions/strategy/reminders 一次计算 |
| 阶段检测 | 6 阶段生命周期（first_scan → maintenance + regression/stagnation）基于 strict_score 轨迹 |
| 策略计算 | union-find 按文件重叠分组为并行 lanes，fixer_leverage 估算自动修复覆盖率 |
| Action 排序 | type_order × impact 双维度排序，auto_fix 优先于 manual_fix |
| 提醒系统 | 15 种上下文提醒 + 3 级优先级 + 计数衰减（threshold=3）+ no_decay 豁免 |
| 多 IDE 适配 | 7 种 IDE skill doc（Claude/Cursor/Codex/Copilot/Windsurf/Gemini/OpenCode）+ 版本化 + overlay 模式 |
| 风险标记 | ignore_suppression + wontfix_gap 双维度风险检测，按 severity 排序 |
最佳实践
1.Structured narrative output for agent consumption
2.Multi-IDE skill doc templating with version tracking
3.Union-find grouping by file overlap for parallel lane detection
4.Phase-aware strategy hints with regression/stagnation special handling
5.Dedicated vs shared file modes for multi-IDE skill doc installation