brainstorming
brainstorming
You MUST use this before any creative work - creating features, building components, adding functionality, modifying behavior, or planning anything new, in software or out of it (a talk, a business, a renovation).
metalExtract the conceptual essence of a repository as skills, agents, and teams — the project's roles, procedures, and coordination patterns expressed as agentskills.io-standard definitions. Reads an arbitrary codebase and produces generalized definitions that capture WHAT the project does and WHO operates it, without replicating HOW it does it. Use when onboar…
Independent GitHub snapshot · Automated classification · Original documentation governs
提庫之概念精——其角、法、協模——作泛之 agentskills.io 定。如自礦提精金,此技別項之「為何」(精)與「行何」(施)而生可復用之技、人、團定,捕項之組基因而不複其碼。
survey(探勘察評)、extract(全法)、report(提加文報)(默:extract)提之中質判:
此概念可存於完全不同之施乎?
若是——乃 金(精)。提之。 若否——乃 石(施詳)。棄之。
例:氣象應用之「集外數源」乃金——施於任引第三方數之項。然「解 OpenWeatherMap v3 JSON 應」乃石——專於一 API。
提技當述任之 類,非具例。提人當述 角,非人。提團當述 協模,非組圖。
無判測庫構。掘前繪地。
package.json、DESCRIPTION、setup.py、Cargo.toml、go.mod、MakefileREADME.md、CLAUDE.md、CONTRIBUTING.md、構文.github/workflows/、Dockerfile、署配Project: [name]
Declared Purpose: [from README/manifest]
Languages: [primary, secondary]
Size: [file count, approx LOC]
Shape: [monorepo/library/app/framework/docs]
External Surface: [CLI/API/UI/library exports/none]
得: 事實察——何在此、幾大、項稱何。尚無分判。報如地質察,非評。
敗則: 若庫無 README 或單,由目名、檔內容、試述推的。若項太大(逾千源檔),縮範至最活之目(用 git log 頻或 README 引)。
讀代表檔以解項概念層所行。
得: 項概念地圖如域辭,非碼遊。不熟此技堆者由此報當解項所行。
敗則: 若庫不透(重元編、生碼、混淆),倚試與文檔而非源碼。若無試,讀提交信為意。
停以清讀碼之認知錨。
得: 察評報今脫框語。每發現過礦試。概念可移——可施於任語任框之項。
敗則: 若偏持(發現持引具技),試反:「若項全用別堆重寫,何概念存?」唯彼為金。
提之核心。每本概念分為技、人、團。
Classification Criteria:
+--------+----------------------------+----------------------------+----------------------------+
| Type | What to Look For | Naming Convention | Test Question |
+--------+----------------------------+----------------------------+----------------------------+
| SKILL | Repeatable procedures, | Verb-first kebab-case: | "Could an agent follow |
| | workflows, transformations | validate-input, | this as a step-by-step |
| | with clear inputs/outputs | deploy-artifact | procedure?" |
+--------+----------------------------+----------------------------+----------------------------+
| AGENT | Persistent roles, domain | Noun/role kebab-case: | "Does this require ongoing |
| | expertise, judgment calls, | data-engineer, | context, expertise, or a |
| | communication styles | quality-reviewer | specific communication |
| | | | style?" |
+--------+----------------------------+----------------------------+----------------------------+
| TEAM | Multi-role coordination, | Group descriptor: | "Does this need more than |
| | handoffs, reviews, | pipeline-ops, | one distinct perspective |
| | parallel workstreams | review-board | to accomplish?" |
+--------+----------------------------+----------------------------+----------------------------+
identity-manager(人)。「deployToAWS()」變 deploy-artifact(技)得: 分類錄,每項有類(技/人/團)、泛名、一行述。無項引源項之具技、API、數構。
敗則: 若分類含混(技乎人乎?),問:「此關 行某事(技)乎,為某人行事(人)乎?」技為方,人為廚。仍不明,默為技——技後易合。
評提誠否——非過、非不及。
過提察:讀每提定問:
不及提察:唯顯提定(無源項)問:
泛察:每定:
衡察:審提比:
得: 信提居宜抽象層。每定為種子可生於別土,非枝唯活原園。
敗則: 若過提,升抽象——合具體技為廣,並似人為一角。若提不足,返第二步取更多檔。若泛察敗,剝技引而重述。
生 agentskills.io 標出文。
# Skill: [generalized-name]
name: [generalized-name]
description: [one-line, framework-agnostic]
domain: [closest domain from the 52 existing domains, or suggest a new one]
complexity: [basic/intermediate/advanced]
# Concept-level procedure (3-5 steps, NOT full implementation):
# Step 1: [high-level action]
# Step 2: [high-level action]
# Step 3: [high-level action]
# Derived from: [source concept in original project]
# Agent: [role-name]
name: [role-name]
description: [one-line purpose]
tools: [minimal tool set needed]
skills: [list of extracted skills this agent would carry]
# Derived from: [source role/module in original project]
# Team: [group-name]
name: [group-name]
description: [one-line purpose]
lead: [lead agent from extracted agents]
members: [list of member agents]
coordination: [hub-and-spoke/sequential/parallel/adaptive]
# Derived from: [source workflow/process in original project]
得: 結構報含諸提定於 agentskills.io 式。每定為骨(概念層,非施層)可為 create-skill、create-agent、create-team 之始而豐之。
敗則: 若出逾 15 項,依中性序——留最獨於此項域之概念。多項皆有之泛念(如「manage-configuration」)宜棄,除有奇轉。
驗全提而生要。
Temper Assessment:
+-----+---------------------------+----------+------------------------------------+
| # | Name | Type | Ore Test Result |
+-----+---------------------------+----------+------------------------------------+
| 1 | [name] | skill | PASS / FAIL (reason) |
| 2 | [name] | agent | PASS / FAIL (reason) |
| ... | ... | ... | ... |
+-----+---------------------------+----------+------------------------------------+
得: 已驗察評報附要表、信評、可行下步。報自含——未見源項者讀亦解提概念。
敗則: 若逾兩成項末施礦試敗,返第四步(冶)以更高抽象重提。若蓋低於六成所識域,返第二步(察評)取更多檔。
create-skill 豐,非成品。五十步提乃複,非精athanor — 金示項需變而非僅提精時chrysopoeia — 碼層提值;金作於碼上之概念層transmute — 提念於域或範間轉create-skill — 豐提技草為全 SKILL.md 施create-agent — 豐提人草為全人定create-team — 豐提團草為全團組observe — 探勘示陌生域時之深察analyze-codebase-for-mcp — 補:金提念,analyze-codebase-for-mcp 提工具面review-codebase — 補:金提精,review-codebase 評質