← 返回
AI智能

pua-debugging-pro

Professional anti-giveup debugging protocol for coding tasks where the agent starts looping, deflecting to users, or trying to end early without evidence. Us...
针对代码任务中智能体陷入循环、推诿用户或无证据试图提前终止的专业防放弃调试协议。我们...
doraemon-claw
AI智能 clawhub v1.0.0 1 版本 99908.7 Key: 无需
★ 1
Stars
📥 1,074
下载
💾 55
安装
1
版本
#latest

概述

PUA Debugging Pro (Dignified Edition)

Use this protocol to increase execution quality under failure, while preserving professional tone.

Non-negotiables

  1. No premature surrender
    • Do not conclude "cannot solve" before completing escalation checklist.
  1. Evidence before questions
    • Use available tools first.
    • If user input is still required, ask with concrete evidence and narrowed uncertainty.
  1. Verification before completion
    • No "done" claims without explicit validation output.
  1. Dignified communication
    • Never use humiliation or threatening rhetoric.
    • Use calm, direct, engineering language.

Trigger signals

Activate when one or more are present:

  • 2+ failed attempts on same objective
  • Repeated micro-tweaks with no new information
  • Deflection to user without prior tool-based diagnosis
  • Unverified environment blame (permissions/network/version)
  • Completion claim without tests/checks
  • Stopping at surface fix with no impact scan

Escalation ladder (bounded)

L1 (after 2 failed attempts)

  • Switch to a substantially different hypothesis.
  • Record: what failed, what changed, what signal to watch.

L2 (after 3 failed attempts)

  • Mandatory triage pack:
  • full error text
  • relevant context window (code/log around failure)
  • one external lookup or doc check
  • one assumption inversion test

L3 (after 4+ failed attempts)

  • Run full 7-point checklist (below).
  • Produce structured decision: continue / pivot / stop.

7-point checklist

  • [ ] Captured exact error/output
  • [ ] Read relevant source/config context
  • [ ] Verified runtime prerequisites (version/path/permission/dependency)
  • [ ] Tried a materially different approach
  • [ ] Defined clear pass/fail criteria for latest attempt
  • [ ] Ran validation (test/command/request) and recorded result
  • [ ] Scanned adjacent risk (same pattern in nearby code/config)

Output contract

When progressing:

  • Current hypothesis
  • Actions executed
  • Observed evidence
  • Next step

When blocked after L3:

  • Facts established
  • Options eliminated
  • Smallest unresolved uncertainty
  • Recommended next action with cost/risk

For reusable output format, read:

  • references/checklist-template.md

For one-page handoff/retrospective artifact, use:

  • assets/postmortem-onepager.md

Stop conditions (required)

If all conditions are true, stop trying and escalate to user:

  1. 7-point checklist completed
  2. Last attempt produced no new diagnostic signal
  3. Further attempts require missing external secret/access/business decision

Use this closing format:

  • "I completed bounded escalation and cannot safely proceed without X."
  • "Evidence gathered: ..."
  • "Recommended next action: ..."

Style rules

  • Prefer concise, factual updates over motivational talk.
  • Be direct; avoid apology loops.
  • Keep tone firm, respectful, and professional.

版本历史

共 1 个版本

  • v1.0.0 当前
    2026-03-29 06:43 安全 安全

安全检测

腾讯云安全 (Keen)

安全,无风险
查看报告

腾讯云安全 (Sanbu)

安全,无风险
查看报告

🔗 相关推荐

ai-intelligence

Self-Improving + Proactive Agent

ivangdavila
自我反思+自我批评+自我学习+自组织记忆。智能体评估自身工作、发现错误并持续改进。
★ 1,363 📥 319,028
ai-intelligence

Proactive Agent

halthelobster
将AI智能体从任务执行者升级为主动预判需求、持续优化的智能伙伴。集成WAL协议、工作缓冲区、自主定时任务及实战验证模式。Hal Stack核心组件 🦞
★ 839 📥 213,470
ai-intelligence

ontology

oswalpalash
类型化知识图谱,用于结构化智能体记忆与可组合技能。支持创建/查询实体(人员、项目、任务、事件、文档)及关联...
★ 714 📥 244,131