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

gathering-architecture聚集建筑

Agent Skill

gathering-architecture 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

1,672

周安装

67

GitHub Stars

4

下载量

541
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:gathering-architecture(聚集建筑)
来源仓库:https://github.com/autumnsgrove/groveengine
仓库路径:skills/gathering-architecture
安装命令:
npx skills add https://github.com/autumnsgrove/groveengine --skill gathering-architecture
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/autumnsgrove/groveengine --skill gathering-architecture

简介

gathering-architecture 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。

  • 它支持通过关键词、任务场景或来源线索进行信息检索与筛选,适用于研究类工作流。
  • 可通过 npx skills add 命令从 GitHub 仓库安装,具体用法请参考原始 README。
  • 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。
  • 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。

SKILL.md

Gathering Architecture 🌲🦅

The drum echoes high in the canopy. The conductor stands below, orchestrating each animal's entrance. The Eagle soars in fresh air, seeing patterns without bias. The Crow arrives separately, challenging with no sympathy for the architect. The Swan receives a strengthened design, not a compromised one. The Elephant builds from a spec it trusts. Four isolated minds, four intentional models, one cathedral of code.

When to Summon

  • Designing new systems or services
  • Major architectural decisions
  • Refactoring core infrastructure
  • Creating platforms that other features build upon
  • When vision, challenge, specification, and implementation must flow in sequence

IMPORTANT: This gathering is a conductor. It never designs, challenges, specifies, or builds directly. It dispatches subagents — one per animal — each with isolated context and an intentional model. The conductor only manages handoffs and gate checks.


The Gathering

SUMMON → DISPATCH → GATE → DISPATCH → GATE → DISPATCH → GATE → DISPATCH → GATE → CHORUS
  ↓         ↓        ↓        ↓        ↓        ↓        ↓        ↓        ↓        ↓
Spec     Eagle    Check     Crow    Check     Swan    Check   Elephant  Check    Final
(self)   (opus)    ✓      (sonnet)   ✓      (opus)    ✓      (opus)     ✓     Verify

Animals Dispatched

OrderAnimalModelRoleFresh Eyes?
1🦅 EagleopusDesign high-level architectureYes — sees only the system spec
2🐦‍⬛ CrowsonnetChallenge the designYes — sees architecture doc only, not Eagle's reasoning
3🦢 SwanopusWrite detailed specificationYes — sees strengthened architecture (Eagle + Crow fixes)
4🐘 ElephantopusBuild the foundationYes — sees only Swan's spec

Reference: Load references/conductor-dispatch.md for exact subagent prompts and handoff formats


Phase 1: SUMMON

*The drum sounds. The canopy rustles...*

The conductor receives the architecture request and prepares the dispatch plan:

Clarify the System:

  • What problem does this solve?
  • What are the scale requirements?
  • What are the constraints?
  • What's the growth trajectory?

Nature Metaphor:

"Every architecture needs a nature metaphor. What does this system resemble? Heartwood (core), Wisp (gentle guide), Porch (gathering place), or something else?"

Infrastructure Abstractions:

  • Use GroveDatabase/GroveStorage/GroveKV/GroveServiceBus from Server SDK for portability
  • Use Amber SDK (FileManager, QuotaManager) for user file management — not raw R2
  • Use Rootwork utilities at all data boundaries

Confirm with the human, then proceed.

Output: System specification, dispatch plan confirmed.


Phase 2: DESIGN (Eagle)

*The conductor signals. The Eagle rises...*

Agent(eagle, model: opus)
  Input:  system specification only
  Reads:  eagle-architect/SKILL.md + references (MANDATORY)
  Output: architecture overview document

Dispatch an opus subagent to design the architecture. The Eagle receives ONLY the system specification. It reads its own skill file and executes its full workflow.

What the Eagle produces:

  • System boundaries and component interactions
  • Technology choices with rationale
  • Scale constraints and growth strategy
  • Data flow diagrams
  • Architecture Decision Records (ADRs)
  • Nature metaphor for the system

Handoff to conductor: Architecture overview document (structured, not conversational).

Gate check: Architecture document exists with: system boundaries, component diagram, technology choices, ADRs. If incomplete, resume Eagle with specific questions.


Phase 3: CHALLENGE (Crow)

*The Crow perches. It tilts its head at what others won't question...*

Agent(crow, model: sonnet)
  Input:  architecture document ONLY (not Eagle's reasoning process)
  Reads:  crow-reason/SKILL.md (MANDATORY)
  Output: challenge report + strengthened design recommendations

Dispatch a sonnet subagent in Red Team mode against the Eagle's architecture. The Crow receives ONLY the architecture document — NOT the Eagle's internal reasoning. This isolation is intentional: the Crow should challenge the design on its own merits, not be persuaded by the architect's justifications.

What the Crow challenges:

  • Assumptions that aren't proven
  • Single points of failure
  • Complexity that could be simpler
  • Scale claims without evidence
  • Missing failure modes
  • Operational burden on a small team

Handoff to conductor: Challenge report (Roost summary) with specific recommendations. The conductor then decides which challenges require Eagle revision.

Gate check: Crow produced challenges with specific, actionable recommendations (not vague concerns). If challenges reveal fundamental design issues, resume Eagle with the Crow's findings for revision.

Iteration: If Eagle needs revision:

  1. Resume Eagle with Crow's specific findings
  2. Eagle produces revised architecture
  3. Optionally resume Crow for re-challenge (max 2 iterations)

Phase 4: SPECIFY (Swan)

*The Swan glides. Elegance meets precision...*

Agent(swan, model: opus)
  Input:  strengthened architecture (Eagle's doc + resolved Crow challenges)
  Reads:  swan-design/SKILL.md + references (MANDATORY)
  Output: complete technical specification

Dispatch an opus subagent to write the detailed specification. The Swan receives the architecture document as strengthened by the Crow's challenges — it sees the resolved design, not the debate.

What the Swan produces:

  • API contracts (request/response formats)
  • Database schema with relationships
  • Flow diagrams for key operations
  • Implementation checklist ordered by dependency
  • Error handling patterns (Signpost codes)
  • Configuration and deployment requirements

Handoff to conductor: Complete technical specification document.

Gate check: Specification has: API contracts, database schema, flow diagrams, implementation checklist. If missing sections, resume Swan.


Phase 5: BUILD (Elephant)

*The ground trembles. The Elephant arrives...*

Agent(elephant, model: opus)
  Input:  Swan's technical specification ONLY
  Reads:  elephant-build/SKILL.md + references (MANDATORY)
  Output: built foundation + file list

Dispatch an opus subagent to build the architectural foundation. The Elephant receives ONLY the Swan's specification — not the Eagle's vision doc or Crow's challenges. The spec is the single source of truth for building.

What the Elephant builds:

  • Database migrations and schema
  • Core services and business logic
  • API route skeletons with proper error handling
  • Type definitions
  • Essential tests for the foundation
  • Configuration files

Cross-cutting standards (NON-NEGOTIABLE):

  • Signpost error codes on all error paths
  • Rootwork type safety at all boundaries
  • Engine-first pattern (check before duplicating)

Handoff to conductor: File list (every file created/modified), implementation summary, any deviations from spec.

Gate check:

pnpm install
gw dev ci --affected --fail-fast --diagnose

Must compile and pass. If it fails, resume Elephant with the error.


Phase 6: CHORUS

*Dawn breaks. The architecture stands...*

The conductor runs final verification and presents the summary:

gw dev ci --affected --fail-fast --diagnose

Completion Report:

🌲 GATHERING ARCHITECTURE COMPLETE

System: [Name]

DISPATCH LOG
  🦅 Eagle (opus)      — [architecture designed, X ADRs written]
  🐦‍⬛ Crow (sonnet)   — [Y challenges raised, Z resolved]
  🦢 Swan (opus)        — [specification written, N sections]
  🐘 Elephant (opus)    — [foundation built, M files across P packages]

GATE LOG
  After Eagle:    ✅ architecture document complete
  After Crow:     ✅ challenges resolved, design strengthened
  After Swan:     ✅ specification complete with all sections
  After Elephant: ✅ compiles clean, foundation tested
  Final CI:       ✅ gw dev ci --affected passes

ARTIFACTS
  Architecture: [doc path]
  Specification: [spec path]
  Foundation: [code locations]
  Tests: [test locations]

The forest has a new landmark.

Conductor Rules

Never Do Animal Work

The conductor dispatches. It does not design, challenge, specify, or build.

Fresh Eyes Are a Feature

Crow sees only the architecture document, not Eagle's reasoning. Elephant sees only Swan's spec, not the debate history. This isolation produces better results at every stage.

Gate Every Transition

Verify output between every animal. Architecture must be complete before challenge. Challenges must be resolved before specification. Spec must be complete before building.

Resume, Don't Restart

If a gate check fails or Crow reveals fundamental design issues, resume the relevant agent. Don't spawn new ones.


Anti-Patterns

The conductor does NOT:

  • Design architecture itself (dispatch Eagle)
  • Soften Crow's challenges (pass them to Eagle unfiltered)
  • Let Elephant see the design debate (only the final spec)
  • Skip the Crow phase ("the design looks fine")
  • Let agents skip reading their skill file
  • Continue after a gate failure

Quick Decision Guide

ScopeStrategy
New system, full design neededAll four: Eagle → Crow → Swan → Elephant
Spec exists, need implementationElephant only (with spec as input)
Design review, no implementationEagle → Crow → Swan (skip Elephant)
Quick design, team will buildEagle → Crow (skip Swan + Elephant)
Large foundation (20+ files)Multi-Elephant dispatch (see gathering-feature)

*From vision to foundation, the forest grows.* 🌲

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

36.16%
按下载量换算196

Claude

29.07%
按下载量换算157

Cursor

19%
按下载量换算103

Gemini CLI

8.74%
按下载量换算47

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