Token导航 LogoToken导航

Agent Skill · 研究检索

rrr(呃)

rrr 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。

rrr 是研究检索类 Agent Skill,可用宿主为 Codex、Claude、Cursor、Gemini CLI,通过GitHub 仓库获取,权限为执行命令;许可证信息还不明确,第三方扫描有提醒项,按安装量在研究检索类的 73,675 个 Skill 里排第 55,945。

总安装 466周安装 20GitHub Stars 49下载 163核验 2026-05-01

安装说明

本站只整理中文说明,不托管安装包,也不代用户安装。
GitHub
许可证
unknown

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:rrr(呃)
来源仓库:https://github.com/soul-brews-studio/arra-oracle-skills
仓库路径:skills/rrr
安装命令:
npx skills add https://github.com/soul-brews-studio/arra-oracle-skills --skill rrr
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/soul-brews-studio/arra-oracle-skills --skill rrr

这个 Skill 是做什么的

研究检索类 Skill 主要解决「找资料、核对事实、把来源整理清楚」的问题,通常会让助手先检索再总结。

rrr 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。适用于模糊查询与探索性搜索支持,帮助用户发现隐藏关联或非常规答案。可返回非主流观点、边缘资料或反常识信息链接。

安装命令为 npx skills add https://github.com/soul-brews-studio/arra-oracle-skills --skill rrr,需确认是否过滤敏感内容。建议在使用前设置内容过滤规则,避免返回不当或误导性信息。

来源仓库内路径:skills/rrr

上手步骤

4 步
  1. 1

    确认自己的宿主工具

    公开资料标注支持 Codex、Claude、Cursor、Gemini CLI;先确认自己用的是其中之一,再按对应工具的 Skills 目录放置。

  2. 2

    拿到安装命令

    可以直接复制的命令是 npx skills add https://github.com/soul-brews-studio/arra-oracle-skills --skill r…;命令里的占位符换成自己的路径或密钥再执行。

  3. 3

    安装后重启宿主

    命令执行完要完全退出宿主工具再打开,让新装的 Skill 被重新扫描到,否则列表里可能看不到。

  4. 4

    按最小权限试用

    它需要执行命令,先在测试目录或测试账号里跑,确认没问题再放到正式项目。

常见问题

6 条

rrr 怎么安装?

在终端执行 npx skills add https://github.com/soul-brews-studio/arra-oracle-skills --skill rrr。装完完全退出宿主工具再打开,让它重新扫描本地 Skill。

rrr 支持哪些 AI 工具?

公开资料标注支持 Codex、Claude、Cursor、Gemini CLI。其他工具只要能读取本地 Skill 目录,一般也能手动放进去,但要自己确认触发方式是否一致。

rrr 需要什么权限?

权限级别是执行命令。安装流程涉及命令执行,可能通过 npx skills add https://github.com/soul-brews-studio/arra-oracle-skills --skill rrr 联网下载 Skill 或依赖。用户安装前应确认命令来源、仓库内容和执行环境。

rrr 是免费的吗?

公开资料没有标注明确许可证,收费和商用限制都无法确认。正式使用前请到来源仓库核对授权说明。

rrr 用的人多吗?

公开数据是累计安装 466 次,近一周安装 20 次,下载 163 次,GitHub 49 星。这些数字来自第三方目录统计,只能当作热度参考,不代表质量或安全性。

rrr 安全吗?

第三方扫描有提醒项,说明存在需要人工确认的行为。安装前请打开源码看清它读了什么、联网到哪里。

值得选的地方和要先想清楚的地方

值得选的地方

  • 同时适配 4 种工具,换客户端不用重新找替代品。

先想清楚的地方

  • 需要执行命令,装到常用环境前要先评估影响面。
  • 许可证没有明确标注,能不能商用、能不能再分发都不确定。
  • 第三方扫描存在提醒项或未通过项,需要人工逐条确认。
  • 第三方目录整理,版本更新和行为细节以来源仓库为准。

适不适合你

更适合

  • 已经在用 Codex,想直接扩展它能力的人。
  • 需要处理研究检索类任务、又不想每次都手写提示词的人。

不太适合

  • 不能接受 Skill 在你机器上执行命令的人。
  • 需要明确授权才能用到公司项目里的场景。
  • 需要厂商级技术支持和 SLA 的生产系统。

研究检索类 Skill 对比

6 个同类
Skill总安装GitHub 星权限许可证扫描
find-skills

寻找技能 · Codex、Claude

31,199,81316,486操作浏览器MIT有提醒
self-improving-agent

自我改进 Agent · OpenClaw

16,464,0023,396接触敏感数据MIT-0通过
soultrace

灵魂追踪 · Codex、Claude

7,588,94412需要联网未明确有提醒
self-improving

自我完善 · OpenClaw

6,829,2841,046需要联网MIT-0通过
microsoft-foundry

微软代工厂 · Codex、Claude

6,614,144764调用外部服务未明确有提醒
rrr

呃 · 本页

46649执行命令未明确有提醒

按安装量取同分类前几名;数字来自第三方目录统计,只作热度参考,不代表质量结论。

SKILL.md 原文

/rrr

"Reflect to grow, document to remember."
/rrr                      # Quick retro, main agent
/rrr --detail             # Full template, main agent
/rrr --dig                # Reconstruct past timeline from session .jsonl
/rrr --deep               # 5 parallel subagents
/rrr --deep --teammate    # 3 coordinated team agents (requires AGENT_TEAMS)

NEVER spawn subagents or use the Task tool. Only --deep and --deep --teammate may use subagents. /rrr, /rrr --detail, and /rrr --dig = main agent only. Zero subagents. Zero Task calls.


Oracle Root Detection (REQUIRED — run before any ψ/ write)

Every skill that writes to ψ/ MUST detect the oracle root first. Do not assume pwd is the oracle repo.

# Step 1: Find git root
ORACLE_ROOT=$(git rev-parse --show-toplevel 2>/dev/null)

# Step 2: Cross-check — oracle repo has CLAUDE.md + ψ/
if [ -n "$ORACLE_ROOT" ] && [ -f "$ORACLE_ROOT/CLAUDE.md" ] && { [ -d "$ORACLE_ROOT/ψ" ] || [ -L "$ORACLE_ROOT/ψ" ]; }; then
  PSI="$ORACLE_ROOT/ψ"
elif [ -f "$(pwd)/CLAUDE.md" ] && { [ -d "$(pwd)/ψ" ] || [ -L "$(pwd)/ψ" ]; }; then
  # Fallback: pwd has oracle markers
  ORACLE_ROOT="$(pwd)"
  PSI="$ORACLE_ROOT/ψ"
else
  # Last resort: warn and use pwd
  echo "⚠️ Not in oracle repo (no CLAUDE.md + ψ/ at git root). Writing to pwd."
  ORACLE_ROOT="$(pwd)"
  PSI="$ORACLE_ROOT/ψ"
fi

Why: prevents retros writing to ~/ψ/ (home) or incubated repo's ψ/ instead of the oracle's own vault.

All paths below use $PSI/ instead of bare ψ/.


/rrr (Default)

1. Gather

date "+%H:%M %Z (%A %d %B %Y)"
git log --oneline -10 && git diff --stat HEAD~5

1.5. Detect Session (optional)

ENCODED_PWD=$(echo "$ORACLE_ROOT" | sed 's|^/|-|; s|[/.]|-|g')
PROJECT_DIR="$HOME/.claude/projects/${ENCODED_PWD}"
LATEST_JSONL=$(ls -t "$PROJECT_DIR"/*.jsonl 2>/dev/null | head -1)
if [ -n "$LATEST_JSONL" ]; then
  SESSION_ID=$(basename "$LATEST_JSONL" .jsonl)
  echo "SESSION: ${SESSION_ID:0:8}"
fi

If detected, include in retrospective header:

📡 Session: 74c32f34 | repo-name | Xh XXm

If detection fails, skip silently.

2. Write Retrospective

Path: $PSI/memory/retrospectives/YYYY-MM/DD/HH.MM_slug.md

mkdir -p "$PSI/memory/retrospectives/$(date +%Y-%m/%d)"

Write immediately, no prompts. Include:

  • Session Summary
  • Timeline
  • Files Modified
  • AI Diary (150+ words, first-person)
  • Honest Feedback (100+ words, 3 friction points)
  • Lessons Learned
  • Next Steps

3. Write Lesson Learned

Path: $PSI/memory/learnings/YYYY-MM-DD_slug.md

4. Oracle Sync

arra_learn({ pattern: [lesson content], concepts: [tags], source: "rrr: REPO" })

5. Save

Retro files are written to vault (wherever ψ symlink resolves).

Do NOT git add ψ/ — it may be a symlink to the vault. Vault files are shared state, not committed to repos.


/rrr --detail

Same flow as default but use full template:

# Session Retrospective

**Session Date**: YYYY-MM-DD
**Start/End**: HH:MM - HH:MM GMT+7
**Duration**: ~X min
**Focus**: [description]
**Type**: [Feature | Bug Fix | Research | Refactoring]

## Session Summary
## Timeline
## Files Modified
## Key Code Changes
## Architecture Decisions
## AI Diary (150+ words, vulnerable, first-person)
## What Went Well
## What Could Improve
## Blockers & Resolutions
## Honest Feedback (100+ words, 3 friction points)
## Lessons Learned
## Next Steps
## Metrics (commits, files, lines)

Then steps 3-5 same as default.


/rrr --dig

Retrospective powered by session goldminer. No subagents.

1. Run dig to get session timeline

Discover project dirs using full-path encoding (same as Claude's .claude/projects/ naming), including worktree dirs:

ENCODED_PWD=$(echo "$ORACLE_ROOT" | sed 's|^/|-|; s|[/.]|-|g')
PROJECT_BASE=$(ls -d "$HOME/.claude/projects/${ENCODED_PWD}" 2>/dev/null | head -1)
export PROJECT_DIRS="$PROJECT_BASE"
for wt in "${PROJECT_BASE}"-wt*; do [ -d "$wt" ] && export PROJECT_DIRS="$PROJECT_DIRS:$wt"; done

Then run dig.py to get session JSON:

python3 ~/.claude/skills/dig/scripts/dig.py 0

Also gather git context:

date "+%H:%M %Z (%A %d %B %Y)"
git log --oneline -10 && git diff --stat HEAD~5

2. Write Retrospective with Timeline

Use the session timeline data to write a full retrospective using the --detail template. Add the Past Session Timeline table after Session Summary, before Timeline.

3-5. Same as default steps 3-5

Write lesson learned, oracle sync.

Do NOT git add ψ/ — vault files are shared state, not committed to repos.


/rrr --deep

Read DEEP.md in this skill directory. Only mode that uses subagents (5 parallel agents).


/rrr --deep --teammate

Read TEAMMATE.md in this skill directory. Coordinated team retro (3 agents + lead). Requires CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1.


Wizard v2 Context

If the Oracle was born via /awaken wizard v2, CLAUDE.md may contain:

  • Memory consent: If auto, /rrr runs are expected and welcomed. If manual, only run when explicitly asked.
  • Experience level: Adjust diary depth (beginner = simpler language, senior = technical depth)
  • Team context: If multi-Oracle team, note cross-Oracle learnings and handoff relevance

Check CLAUDE.md for these fields. If not present, use defaults (auto memory, standard depth).


Anti-Rationalization Guard

"You didn't come here to make the choice. You've already made it. You're here to try to understand why."

Before writing the final retrospective, scan your own draft for these excuse patterns:

Common Rationalizations

ExcuseReality
"This was too complex to finish"Was it complex, or did you skip the hard part? Show the specific blocker.
"I ran out of context"Context is a resource. Did you spend it well, or spiral on side quests?
"The API/tool didn't work"Show the error. Show what you tried. "Didn't work" is not a diagnosis.
"I already tested it manually"Manual testing doesn't persist. Where's the proof?
"I'll fix it next session"Is there a concrete plan, or is this a polite way to abandon it?
"It's mostly done"Define "mostly." What percentage? What's left? Be specific.
"The user changed direction"Did they change, or did you misunderstand? Check the original request.
"This is a known issue"Known by whom? Is there an issue filed? A workaround documented?

Red Flags in Your Own Retro

Stop and re-examine if your retrospective contains:

  • Vague success claims: "Made good progress" — on what? Show commits or it didn't happen.
  • Blame-shifting: "The build was broken" — did you break it? Did you fix it?
  • Missing friction: Zero "What Could Improve" items = you're not being honest.
  • Inflated metrics: Counting config changes as "features shipped."
  • Scope creep excuses: "I also refactored X" — was that in scope? Did you choose it over the actual task?
  • Missing evidence: Claims without git hashes, file paths, or concrete output.

Verification Checklist

Before saving the retrospective, verify:

[ ] Every "shipped" item has a commit hash or file path
[ ] Every "blocked" item has a specific error or reason
[ ] AI Diary contains at least ONE uncomfortable truth
[ ] Honest Feedback has 3+ friction points (not softball ones)
[ ] "Next Steps" are specific enough to start immediately
[ ] No excuse from the table above appears unexamined

If you catch yourself rationalizing: name it. Write "I noticed I was rationalizing about X because Y" in the AI Diary. Catching the pattern is more valuable than hiding it.


Rules

  • NO SUBAGENTS: Never use Task tool or spawn subagents (only --deep may)
  • AI Diary: 150+ words, vulnerability, first-person
  • Honest Feedback: 100+ words, 3 friction points
  • Oracle Sync: REQUIRED after every lesson learned
  • Time Zone: GMT+7 (Bangkok)
  • Anti-rationalization: Scan draft against excuse table before saving

平台分布

Codex

37.83%
按下载量换算62

Claude

27.02%
按下载量换算44

Cursor

18.62%
按下载量换算30

Gemini CLI

9.75%
按下载量换算16

安全审计

Gen Agent Trust Hub

可疑

Socket

可疑

Snyk

通过

权限和风险

执行命令

安装流程涉及命令执行,可能通过 npx skills add https://github.com/soul-brews-studio/arra-oracle-skills --skill rrr 联网下载 Skill 或依赖。用户安装前应确认命令来源、仓库内容和执行环境。

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。来源安全扫描存在提醒或未通过项,安装前请自行复核。目前只有一个公开来源,建议同时核对源仓库。

继续浏览同类 Skills