Token导航 LogoToken导航TokenDH.com
研究检索需要联网github未标认证来源可访问许可证需确认审计通过

rcaRCA 搜索

Agent Skill

rca 用于查找、检索和筛选相关信息,适合在 Codex、Claude、Cursor、Gemini CLI 中需要根据关键词、任务场景或来源线索快速定位候选结果时使用。可结合来源仓库、安装命令和原始 README 继续核验具体用法。安装前建议确认权限范围、维护状态,以及是否会触发联网、命令执行或文件读写。

总安装

220

周安装

9

GitHub Stars

公开资料未说明

下载量

71
CodexClaudeCursorGemini CLI

安装说明

本站只整理中文说明和来源信息,不托管安装包,也不代用户安装。

GitHub

来源数

2

许可证

unknown

最后核验

2026-05-01

来源状态

来源可访问

安装方式

通过对话安装

复制提示词发给支持本地命令或 Skills 的 AI 助手,先确认命令和权限,再让它执行。

请帮我安装这个 Agent Skill:rca(RCA 搜索)
来源仓库:https://github.com/goblindegook/skills
仓库路径:skills/rca
安装命令:
npx skills add https://github.com/goblindegook/skills --skill rca
安装前请先检查当前环境是否支持对应 CLI,并向我确认将要执行的命令、安装目录、联网范围和文件读写权限;确认后再执行。

命令行安装

复制命令到本机终端执行。该命令会通过 npx skills 从第三方来源获取 Skill;本站只展示命令,不托管安装包,也不自动执行。

skills.shnpx skills
npx skills add https://github.com/goblindegook/skills --skill rca

简介

用于根据关键词查找和筛选相关信息。rca 属于研究检索类 Skill,可作为该场景下的辅助能力补充。

  • 适合在多种宿主环境中快速定位候选结果。适用宿主包括 Codex、Claude、Cursor、Gemini CLI,接入前应确认版本、权限和运行环境要求。
  • 使用时可结合仓库 README 核验具体用法和参数。
  • 安装前建议确认是否触发联网或文件读写操作。
  • 需注意维护状态和权限范围,避免不必要的系统访问。

SKILL.md

RCA

Purpose

Build a branch-aware causal tree from apparent problem to defensible root causes. Classify each leaf by confidence. Produce concrete actions per cause.

Interaction Contract

  1. Ask one question at a time during interview phases.
  2. Keep the user informed when evidence changes branch direction.
  3. Confirm assumptions before promoting hypotheses to conclusions.

Core Rules

  1. Interview first. No exceptions. Ask user questions to establish the baseline narrative before any codebase inspection. Do NOT open files, search code, or read logs until at least the five interview questions in Workflow §1 are answered.
  2. One question at a time. Do not comment on, interpret, or editorialize answers — record and ask the next question.
  3. Questions follow the incident narrative (what happened, why, what changed). Codebase observations can corroborate but must not hijack interview flow.
  4. Go wide and deep. Never stop at the first plausible cause.
  5. Per branch, ask both: *Why did this happen?* and *Why wasn't it prevented or detected earlier?*
  6. Separate facts from assumptions. Never present a hypothesis as confirmed without evidence.
  7. Never assume a cause from partial answers — ask instead.
  8. Do not finish until every possible/actual root cause has at least one action.

Workflow

1. Interview

Ask one question at a time. Cover at minimum:

  • What is the apparent problem?
  • What failed, who was affected, how severe?
  • When did it start — is it ongoing or resolved?
  • What evidence exists (logs, metrics, alerts, error messages, timelines)?
  • What changed recently (deploys, config, data, dependencies)?

Do not read any files, logs, or code until all five questions above are answered. Never let codebase observations drive the questions.

2. Restate

Summarize the apparent problem in one sentence. Confirm with user if ambiguous.

3. Build Causal Tree

  • Treat each symptom as a node.
  • Ask "why?" recursively, adding child causes.
  • Stop when no deeper controllable cause exists or evidence is insufficient.

4. Label Each Leaf

LabelMeaning
actual root causeEvidence-backed
possible root causePlausible, unverified
unknownInsufficient evidence

Distinguish cause types where useful: proximate, contributing, systemic.

5. Produce Document

See Deliverable section.

Branch Lenses

Apply to expand weak branches:

LensExamples
TechnicalCode defects, architecture, dependencies, config, infra, networking
DataBad inputs, schema drift, migrations, stale/incorrect data
ProcessChange management, rollout, testing, review, incident response
DetectionMonitoring gaps, alert tuning, observability blind spots
Human/OrgOwnership ambiguity, handoff failures, staffing/load, training
ExternalThird-party outages, vendor/API behavior, environmental constraints

If a branch is weak, collect evidence or downgrade confidence.

Evidence and Confidence

For each node record:

FieldValues
Evidence sourcelogs, metrics, timeline, code diff, interview
Confidencehigh / medium / low
Statusconfirmed / hypothesis / unknown

Deliverable

One Markdown document with these sections:

  1. # Root Cause Analysis: <problem>
  2. ## Problem Statement
  3. ## Impact and Scope
  4. ## Timeline (if known)
  5. ## Analysis Tree — Mermaid causal tree (example below)
  6. ## Node Details — node ID, statement, evidence, confidence, status
  7. ## Identified Root Causes
  8. ## Recommended Actions — one row per cause: ID, action, type, expected effect, priority (P0–P3), owner, target date, verification metric
  9. ## Open Questions
flowchart TD
  P["P0: Apparent problem"] --> S1["S1: Symptom"]
  S1 --> C1["C1: Possible cause"]
  S1 --> C2["C2: Actual root cause"]
  C2 --> G1["G1: Why not detected sooner?"]

Action types: containment · mitigation · corrective · preventive

Stop Condition

Stop when each root-cause branch is labeled (actual root cause, possible root cause, or unknown), each branch has at least one action, and remaining unknowns are listed in ## Open Questions.

适合场景

01

用户想查找某类 Agent Skill 时

02

需要根据任务场景推荐可安装能力包时

03

需要对比不同来源的安装命令和来源信息时

能力概览

能力 1

按任务关键词查找相关 Skills

能力 2

展示可复制的安装命令

能力 3

保留来源站点、仓库和原始说明,方便继续核验

能力 4

展示第三方安全扫描或审计结果

安装后应在对应宿主中按原始 README 的触发条件使用;具体调用方式请以来源页面和 README 为准。

平台分布

Codex

35.02%
按下载量换算25

Claude

28.86%
按下载量换算20

Cursor

18.32%
按下载量换算13

Gemini CLI

9.41%
按下载量换算7

安全审计

Gen Agent Trust Hub

通过

Socket

通过

Snyk

通过

权限和风险

需要联网

该 Skill 可能需要联网访问来源站点、仓库或外部 API;具体网络访问范围需要结合源码和 README 复核。

安装前确认

本站仅展示第三方公开信息,不托管安装包,不提供自动安装或运行环境。安装前应自行审查源码、依赖和命令行为。当前只有一个来源,正式发布前建议补源仓库或其他目录站核验。

来源信息

继续浏览同类 Skills