원클릭으로 Manus에서 모든 스킬 실행

crash-cluster-triage

스타3

포크1

업데이트2026년 6월 1일 01:48

Capture a WER dump, root-cause a reproducible WoW.exe (FG) crash, and ship a hardening patch + regression guard, documented as a crash cluster. Use when the foreground client crashes repeatably.

설치

Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.

Manus에서 실행

출처

QiMata

QiMata/BloogBot

GitHub 저장소 열기 Creator 저장소 보기

다운로드

Manus에서 실행

Crash Cluster Triage

Goal

Turn a reproducible foreground-client crash into a documented cluster: capture the dump, root-cause it, ship a hardening patch with a regression guard, and record the cluster so it doesn't recur silently.

Inputs

A reproducible crash (console stderr, WoW.exe unhandled exception, or WER report) and its trigger sequence.
Key references:
- Crash research home: docs/Plan/Crashes/README.md (structure + WER capture plan); historical reference: CRASH-001 (a known client bug — an acceptable test skip per CLAUDE.md).
- Hardening sites: Exports/BotRunner/ (FG guards), Services/WoWStateManager/ (state guards), native Navigation.dll (collision edge cases).
Area rules: .github/instructions/services.instructions.md, .github/instructions/native.instructions.md.

Status: The crash-research structure is defined in docs/Plan/Crashes/README.md, but WER-capture tooling is not yet implemented on this branch — set up local dumps manually (below) per cluster.

Preconditions

A minimal, repeatable reproduction (set, GM commands, task sequence).
Process safety: only kill the specific WoW.exe PID you launched — never blanket-kill (D2Bot and other repos share this machine).

Procedure

Create a cluster doc docs/Plan/Crashes/crash-<name>.md with Symptoms and Reproduction.
Enable local dumps: registry HKLM\SOFTWARE\Microsoft\Windows\Windows Error Reporting\LocalDumps\WoW.exe → DumpFolder, DumpType=2 (full), DumpCount.
Reproduce, collect the dump, and root-cause it (WinDbg) — record the faulting stack in the cluster doc.
Ship a hardening patch at the right layer (null/bounds guard, state-machine fix, native edge-case fix) + a regression test or guard.
Add a FailureReason for the path if appropriate (see [[failure-reason-mapping]]), and mark the cluster Hardened.

Verification

The reproduction no longer crashes after the patch.
The regression test/guard is green; broader suite unaffected (.\scripts\test-fast.ps1).
The cluster doc records symptoms → root cause → fix → guard.

Outputs

docs/Plan/Crashes/crash-<name>.md, a hardening patch, and a regression guard.

Failure modes and recovery

Patching a symptom, not the cause — confirm the faulting stack from the dump.
Blanket process kills during repro — kill only your PID.
No regression guard — the cluster will recur; always add one (or document why it's an accepted client bug like CRASH-001).

Related skills

[[failure-reason-mapping]] — classify the crash path.
[[fg-bg-physics-parity]] — physics edge cases that crash the client.
[[debugging]] — general investigation workflow.
Reference: docs/Plan/Crashes/README.md.

이 저장소의 다른 Skills

같은 저장소

debugging

QiMata/BloogBot

Debugging workflow for BloogBot/WWoW codebase. Use when investigating bugs, errors, unexpected behavior, or service failures.

2026-06-023

activity-catalog-bootstrap

QiMata/BloogBot

Add or generate compiled activity-catalog rows (ActivityDefinition) from the leveling-guide source and keep them passing the catalog invariant tests + MaNGOS cross-validation. Use when adding an Activity to the catalog.

2026-06-013

bot-profile

QiMata/BloogBot

Creating or modifying WoW class/spec combat profiles. Use when adding a new class profile, fixing combat rotations, or adjusting bot behavior for a specific specialization.

2026-06-013

botrunner-task-implementation

QiMata/BloogBot

Add an IBotTask behavior-tree task that orchestrates Actions to drive one minute state change, with FG+BG parity, tests, and live-validation. Use when implementing a new bot behavior (move/loot/interact/cast/gather/etc.) as a Task on the LIFO task stack.

2026-06-013

config-hot-reload-subscriber

QiMata/BloogBot

Add an IConfigSubscriber for a new reloadable config section so a running service applies changes without restart, with ACK/rollback. Use when a config section must hot-reload (the feature-flag / live-tunable path).

2026-06-013

coordinator-implementation

QiMata/BloogBot

Author a new activity coordinator (Dungeon/Raid/BG/Quest/Economy/etc.) in WoWStateManager that orchestrates multiple bots through an Activity by emitting Objectives from snapshot state. Use when a multi-character Activity needs server-side coordination.

2026-06-013

name	crash-cluster-triage
description	Capture a WER dump, root-cause a reproducible WoW.exe (FG) crash, and ship a hardening patch + regression guard, documented as a crash cluster. Use when the foreground client crashes repeatably.
trigger	crash triage, WoW.exe crash, WER dump, FG crash, root cause a crash, hardening patch, crash cluster, CRASH-001

Crash Cluster Triage

Goal

Inputs

A reproducible crash (console stderr, WoW.exe unhandled exception, or WER report) and its trigger sequence.
Key references:
- Crash research home: docs/Plan/Crashes/README.md (structure + WER capture plan); historical reference: CRASH-001 (a known client bug — an acceptable test skip per CLAUDE.md).
- Hardening sites: Exports/BotRunner/ (FG guards), Services/WoWStateManager/ (state guards), native Navigation.dll (collision edge cases).
Area rules: .github/instructions/services.instructions.md, .github/instructions/native.instructions.md.

Status: The crash-research structure is defined in docs/Plan/Crashes/README.md, but WER-capture tooling is not yet implemented on this branch — set up local dumps manually (below) per cluster.

Preconditions

A minimal, repeatable reproduction (set, GM commands, task sequence).
Process safety: only kill the specific WoW.exe PID you launched — never blanket-kill (D2Bot and other repos share this machine).

Procedure

Create a cluster doc docs/Plan/Crashes/crash-<name>.md with Symptoms and Reproduction.
Enable local dumps: registry HKLM\SOFTWARE\Microsoft\Windows\Windows Error Reporting\LocalDumps\WoW.exe → DumpFolder, DumpType=2 (full), DumpCount.
Reproduce, collect the dump, and root-cause it (WinDbg) — record the faulting stack in the cluster doc.
Ship a hardening patch at the right layer (null/bounds guard, state-machine fix, native edge-case fix) + a regression test or guard.
Add a FailureReason for the path if appropriate (see [[failure-reason-mapping]]), and mark the cluster Hardened.

Verification

The reproduction no longer crashes after the patch.
The regression test/guard is green; broader suite unaffected (.\scripts\test-fast.ps1).
The cluster doc records symptoms → root cause → fix → guard.

Outputs

docs/Plan/Crashes/crash-<name>.md, a hardening patch, and a regression guard.

Failure modes and recovery

Patching a symptom, not the cause — confirm the faulting stack from the dump.
Blanket process kills during repro — kill only your PID.
No regression guard — the cluster will recur; always add one (or document why it's an accepted client bug like CRASH-001).

Related skills

[[failure-reason-mapping]] — classify the crash path.
[[fg-bg-physics-parity]] — physics edge cases that crash the client.
[[debugging]] — general investigation workflow.
Reference: docs/Plan/Crashes/README.md.