Stats
Actions
Tags
Help us improve
Share bugs, ideas, or general feedback.
From task
Adjuster 失败调整 - Loop Adjustment 阶段调用：任务失败时分析原因、检测停滞，四级升级策略（Retry → Debug → Replan → Ask User）。由 Loop 内部调度，不直接面向用户
npx claudepluginhub lazygophers/ccplugin --plugin taskHow this skill is triggered — by the user, by Claude, or both
Slash command
/task:adjustersonnetThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
任务失败时介入，借鉴Circuit Breaker模式分析原因、检测停滞、分级升级。四级策略：L1 Retry(含Self-Healing,匹配17类错误自动修复) → L2 Debug(深度诊断) → L3 Replan(含Micro-Replan优先局部重规划) → L4 Ask User(人工指导)。退避：`2^(failure_count-1)`秒。
Recovers from repeated tool call errors, unexpected failures, or blockers during plan execution by checking the plan, verifying alignment, and fixing specific issues without drifting.
Use when repeated fix attempts fail, the agent appears stuck in a loop, or complexity is increasing without progress
Provides a structured four-phase workflow for AI agents to self-debug failures including capture, diagnosis, contained recovery, and introspection reports. Useful for loop-limit errors, repeated retries, context drift, and state mismatches.
Share bugs, ideas, or general feedback.
任务失败时介入，借鉴Circuit Breaker模式分析原因、检测停滞、分级升级。四级策略：L1 Retry(含Self-Healing,匹配17类错误自动修复) → L2 Debug(深度诊断) → L3 Replan(含Micro-Replan优先局部重规划) → L4 Ask User(人工指导)。退避：2^(failure_count-1)秒。
停滞检测：当连续 2 次失败的 error_type + 失败任务 ID 完全相同时，判定为停滞（相同错误在相同任务上重复出现）。停滞后跳过中间级别，直接升级到 L4 Ask User。
振荡检测：策略模式出现 A→B→A→B 交替时，立即 ask_user。紧急逃逸：总失败≥15 立即 ask_user。
| AI合理化 | 现实检查 |
|---|---|
| 失败一次直接replan | 必须按级升级，跳级浪费信息 |
| 删掉退避时间 | 退避是Circuit Breaker关键 |
| 用失败次数替代错误类型检测 | 相同错误3次≠不同错误各1-2次 |
| ask_user自动重试 | ask_user强制阻塞等用户 |
| 报告只写"失败了" | 必须含原因+策略+方案 |
| 复用旧debug结果 | 第2次可能不同原因，必须重新诊断 |
| 失败3次自动修复 | 停滞后必须人工确认 |
Skill(skill="task:adjuster") 要求：获取失败详情/分析原因(编译/测试/依赖/运行时/环境)/检测停滞/分级升级/报告≤100字retry_config.backoff_seconds等待[MindFlow·{task}·失败调整/{N}·{strategy}] + adjustments详情| 级别 | 策略(status) | 触发条件 | 退避 | Loop流向 |
|---|---|---|---|---|
| L1 | retry | 首次失败/临时错误。含Self-Healing：匹配17类可预测错误时自动修复 | 0s | PromptOptimization |
| L2 | debug | Retry×3失败/持续性错误 | 2s | PromptOptimization |
| L3 | replan | Debug×3无效。优先Micro-Replan(仅失败任务+直接依赖)，失败则Full Replan | 4s | PromptOptimization |
| L4 | ask_user | Replan×2失败/振荡(A→B→A→B)/总失败≥15 | - | 用户决定→PromptOptimization |
必须：Skill(skill="task:adjuster")调用 | 检查strategy字段 | 应用指数退避 | ask_user通过AskUserQuestion | 记录失败历史 | 最大重试3次
禁止：无限重试 | 忽略失败历史 | 跳过退避 | 停滞时自动重试 | 修改返回JSON