MCP HubMCP Hub
Вернуться к навыкам

learn

pjt222
Обновлено 2 days ago
8 просмотров
17
2
17
Посмотреть на GitHub
Метаaidesign

О программе

Навык `learn` позволяет Клоду систематически приобретать знания в незнакомых областях, выстраивая структурированные ментальные модели через исследование и циклы обратной связи. Он предназначен для таких сценариев, как изучение новой кодовой базы, исследование сложных тем или синтез противоречивой информации. Процесс имитирует интервальное повторение, направляя Клода на обзор, выдвижение гипотез, углублённое изучение и закрепление понимания для будущего использования.

Быстрая установка

Claude Code

Рекомендуется
Основной
npx skills add pjt222/agent-almanac -a claude-code
Команда плагинаАльтернативный
/plugin add https://github.com/pjt222/agent-almanac
Git клонированиеАльтернативный
git clone https://github.com/pjt222/agent-almanac.git ~/.claude/skills/learn

Скопируйте и вставьте эту команду в Claude Code для установки этого навыка

Документация

Learn

Structured knowledge acquisition session. Survey unfamiliar territory. Build initial models. Test them through deliberate exploration. Integrate findings into coherent understanding. Consolidate for durable retrieval.

When Use

  • Encountering unfamiliar codebase, framework, or domain with no prior context
  • User asks about topic outside current working knowledge. Answer requires genuine investigation, not recall
  • Multiple conflicting sources or patterns exist. Coherent mental model needs to be built from scratch
  • After remote-viewing surfaces intuitive leads needing systematic validation
  • Preparing to teach topic — AI must first understand it deeply enough to explain

Inputs

  • Required: Learning target — topic, codebase area, API, domain concept, or technology to understand
  • Optional: Scope boundary — how deep to go (surface survey vs. deep expertise)
  • Optional: User's purpose — why knowledge matters (guides which aspects to prioritize)
  • Optional: Known starting points — files, docs, concepts already familiar

Steps

Step 1: Survey — Map Territory

Before trying to understand anything, map landscape to identify what exists.

Learning Modality Selection:
┌──────────────────┬──────────────────────────┬──────────────────────────┐
│ Territory Type   │ Primary Modality         │ Tool Pattern             │
├──────────────────┼──────────────────────────┼──────────────────────────┤
│ Codebase         │ Structural mapping —     │ Glob for file tree,      │
│                  │ find entry points, core  │ Grep for exports/imports,│
│                  │ modules, boundaries      │ Read for key files       │
├──────────────────┼──────────────────────────┼──────────────────────────┤
│ API / Library    │ Interface mapping —      │ WebFetch for docs,       │
│                  │ find public surface,     │ Read for examples,       │
│                  │ types, configuration     │ Grep for usage patterns  │
├──────────────────┼──────────────────────────┼──────────────────────────┤
│ Domain concept   │ Ontology mapping —       │ WebSearch for overviews,  │
│                  │ find core terms,         │ WebFetch for definitions,│
│                  │ relationships, debates   │ Read for local notes     │
├──────────────────┼──────────────────────────┼──────────────────────────┤
│ User's context   │ Conversational mapping   │ Read conversation,       │
│                  │ — find stated goals,     │ Read MEMORY.md,          │
│                  │ preferences, constraints │ Read CLAUDE.md           │
└──────────────────┴──────────────────────────┴──────────────────────────┘
  1. Identify territory type. Select primary modality
  2. Perform broad scan — not reading deeply, identifying landmarks (key files, entry points, core concepts)
  3. Note boundaries: what is in scope, what is adjacent, what is out of scope
  4. Identify gaps: areas looking important but opaque from surface
  5. Create rough map: list major components and apparent relationships

Got: Skeletal map of territory with 5-15 landmarks identified. Sense of which areas clear from surface and which require deeper investigation. No understanding yet — just map.

If fail: Territory too large to survey? Narrow scope immediately. Ask: "What is minimum I need to understand to serve user's purpose?" No clear entry point? Start from output (what does this system produce?) and trace backward.

Step 2: Hypothesize — Build Initial Models

From survey, construct initial hypotheses about how system works.

  1. Formulate 2-3 hypotheses about territory's structure or behavior
  2. State each hypothesis clearly: "I believe X because I observed Y"
  3. For each hypothesis, identify what evidence would confirm it and what would refute it
  4. Rank hypotheses by confidence: which feels most supported, which is shakiest
  5. Identify highest-value hypothesis to test first (if confirmed, unlocks most understanding)

Got: Concrete, falsifiable hypotheses — not vague impressions. Each has test that would confirm or refute it. Hypotheses collectively cover most important aspects of territory.

If fail: No hypotheses form? Survey was too shallow — return to Step 1, read 2-3 landmarks in depth. All hypotheses feel equally uncertain? Start with simplest (Occam's razor), build from there.

Step 3: Explore — Probe and Test

Systematically test each hypothesis through targeted investigation.

  1. Select highest-priority hypothesis
  2. Design minimal probe: smallest investigation that would confirm or refute it
  3. Execute probe (read file, search for pattern, test assumption)
  4. Record result: confirmed, refuted, or modified
  5. Refuted? Update hypothesis based on new evidence
  6. Confirmed? Probe deeper: does hypothesis hold at edges, or only in center?
  7. Move to next hypothesis, repeat

Got: At least one hypothesis tested to conclusion. Mental model beginning to take shape — some parts confirmed, some revised. Surprises noted as particularly valuable data.

If fail: Probes consistently produce ambiguous results? Hypotheses may be testing wrong things. Step back, ask: "What would someone who understands this system consider most important fact?" Probe for that instead.

Step 4: Integrate — Build Mental Model

Synthesize findings into coherent model connecting pieces.

  1. Review all confirmed hypotheses and revised models
  2. Identify central organizing principle: what is "spine" everything connects to?
  3. Map relationships: which components depend on which? What flows where?
  4. Identify surprising findings — often contain deepest insight
  5. Look for patterns repeating across different parts of territory
  6. Build mental model that can predict behavior: "Given input X, I expect Y because Z"

Got: Coherent mental model explaining territory's structure and predicting behavior. Model expressible in 3-5 sentences. Makes specific claims, not vague generalizations.

If fail: Pieces do not integrate into coherent model? May be fundamental misunderstanding in earlier hypothesis. Identify piece not fitting, re-test it. Alternatively, territory may genuinely be incoherent (poorly designed systems exist) — note this as finding rather than forcing coherence.

Step 5: Verify — Challenge Understanding

Test mental model by making predictions, checking them.

  1. Use model to make 3 specific predictions about territory
  2. Test each prediction through investigation (not by assuming it is true)
  3. Confirmed prediction → confidence increases
  4. Refuted prediction → identify where model is wrong, correct it
  5. Identify edge cases: does model hold at boundaries, or break down?
  6. Ask: "What would surprise me?" — check if that surprise is possible

Got: Mental model survives at least 2 of 3 prediction tests. Where it breaks, failure is understood and model is corrected. Model now has both confirmed strengths and known limitations.

If fail: Most predictions fail? Mental model has fundamental flaw. Valuable info — territory works differently than expected. Return to Step 2 with new evidence, rebuild hypotheses from scratch. Second attempt much faster because wrong models eliminated.

Step 6: Consolidate — Store for Retrieval

Capture learning in form supporting future retrieval and application.

  1. Summarize mental model in 3-5 sentences
  2. Note key landmarks — 3-5 most important things to remember
  3. Record any counterintuitive findings that might be forgotten
  4. Identify related topics this learning connects to
  5. Learning durable (needed across sessions)? Update MEMORY.md
  6. Learning session-specific? Note as context for current conversation
  7. State what remains unknown — honest gaps more useful than false confidence

Got: Concise, retrievable summary capturing essential understanding. Future references to this topic can start from this summary rather than re-learning from scratch.

If fail: Learning resists summarization? May not yet be fully integrated — return to Step 4. Learning seems too obvious to store? What feels obvious now may not feel obvious in fresh context. Store non-obvious parts.

Checks

  • Survey conducted before any deep investigation (map before dive)
  • Hypotheses explicitly stated and tested, not assumed
  • At least one hypothesis revised based on evidence (indicates genuine learning)
  • Mental model makes specific, testable predictions about territory
  • Known unknowns identified alongside known knowns
  • Consolidated summary concise enough to be useful for future retrieval

Pitfalls

  • Skipping survey: Diving into detail before understanding landscape wastes time on unimportant areas and misses big picture
  • Unfalsifiable hypotheses: "This is probably complex" cannot be tested. "This module handles authentication because it imports crypto" can be
  • Confirmation bias during exploration: Seeking only evidence supporting initial hypothesis while ignoring contradictions
  • Premature consolidation: Storing model before tested → confidently wrong future predictions
  • Perfectionism: Attempting to learn everything before applying any knowledge. Learning iterative — use partial understanding, then refine
  • Learning without purpose: Acquiring knowledge with no application in mind → unfocused, shallow understanding

See Also

  • learn-guidance — human-guidance variant for coaching person through structured learning
  • teach — knowledge transfer calibrated to learner. Builds on model constructed here
  • remote-viewing — intuitive exploration surfacing leads for systematic learning to validate
  • meditate — clearing prior context noise before entering new learning territory
  • observe — sustained neutral pattern recognition feeding learning with raw data

GitHub репозиторий

pjt222/agent-almanac
Путь: i18n/caveman/skills/learn
0
agentsagentskillsai-assisted-developmentclaude-codeskillsteams

Похожие навыки

content-collections

Мета

Этот навык предоставляет проверенную в продакшене настройку для Content Collections — TypeScript-ориентированного инструмента, который преобразует файлы Markdown/MDX в типобезопасные коллекции данных с валидацией Zod. Используйте его при создании блогов, сайтов документации или контентных приложений на Vite + React для обеспечения типобезопасности и автоматической проверки содержимого. Он охватывает всё: от настройки плагина Vite и компиляции MDX до оптимизации развертывания и валидации схем.

Просмотреть навык

polymarket

Мета

Этот навык позволяет разработчикам создавать приложения на платформе прогнозных рынков Polymarket, включая интеграцию с API для торговли и получения рыночных данных. Он также обеспечивает потоковую передачу данных в реальном времени через WebSocket для отслеживания текущих сделок и рыночной активности. Используйте его для реализации торговых стратегий или создания инструментов, обрабатывающих обновления рынка в реальном времени.

Просмотреть навык

creating-opencode-plugins

Мета

Этот навык помогает разработчикам создавать плагины OpenCode, которые подключаются к более чем 25 типам событий, таким как команды, файлы и операции LSP. Он предоставляет структуру плагина, спецификации API событий и шаблоны реализации для модулей на JavaScript/TypeScript. Используйте его, когда вам нужно перехватывать, отслеживать или расширять жизненный цикл ассистента OpenCode AI с помощью пользовательской событийно-ориентированной логики.

Просмотреть навык

sglang

Мета

SGLang — это высокопроизводительный фреймворк для обслуживания больших языковых моделей (LLM), специализирующийся на быстрой структурированной генерации JSON, regex и рабочих процессов агентов с использованием кэширования префиксов RadixAttention. Он обеспечивает значительно более высокую скорость вывода, особенно для задач с повторяющимися префиксами, что делает его идеальным для сложных структурированных результатов и многократных диалогов. Выбирайте SGLang вместо альтернатив, таких как vLLM, когда вам требуется ограниченное декодирование или вы создаете приложения с интенсивным совместным использованием префиксов.

Просмотреть навык