메뉴
기본적으로 소스를 먼저 확인하는 Prompt가 선택됩니다. 직접 명령으로 전환하거나 로컬 사본을 다운로드할 수도 있습니다.
설치 여부를 결정하기 전에 SKILL.md와 SkillsMP에 표시된 보조 파일을 읽어 보세요.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
직접 명령은 검토 Prompt를 거치지 않습니다. 실행하기 전에 소스를 확인하세요.
npx skills add https://github.com/majiayu000/claude-skill-registry --skill adventure명령은 한 줄로 유지됩니다. 복사하기 전에 가로로 스크롤해 전체 내용을 확인하세요.
로컬 사본을 원하시나요? SkillsMP에서 현재 제공할 수 있는 파일을 다운로드하세요.
LLM token logprobs and calibration. Per-decision confidence, ECE, Brier, reliability diagrams, low-confidence triage.
Analyze LLM token logprobs and calibration. Use for per-decision confidence, ECE, Brier scores, reliability diagrams, and low-confidence triage.
回顾最近 N 天的 Claude Code 使用记录——扫描原始会话数据，按主题分组汇总"我都做了什么"，并从个人操作系统视角输出模式、风险与增删建议。当用户说 /recap、"看看我这几天做了什么"、"回顾一下我最近的会话"、"这两天我用 claude 干了啥"、"活动回顾" 时使用。
SOC 직업 분류 기준
SKILL.md 표시 중
| name | adventure |
| description | Room-based exploration with narrative evidence collection |
| allowed-tools | ["read_file","write_file","list_dir"] |
| tier | 1 |
| protocol | ADVENTURE |
| lineage | Colossal Cave, Zork, MUD, LambdaMOO |
| inherits | simulation |
| related | ["room","character","incarnation","simulation","card","memory-palace","world-generation","debugging","sniffable-python"] |
| tags | ["moollm","exploration","narrative","investigation","game","interactive-fiction"] |
| templates | [{"file":"ADVENTURE.yml.tmpl","purpose":"Complete adventure state (inherits simulation properties)"},{"file":"LOG.md.tmpl","purpose":"Summary table (turns, locations, files changed)"},{"file":"TRANSCRIPT.md.tmpl","purpose":"Pure narration (story, YAML objects, mermaid diagrams)"}] |
"Every directory is a room. Every file is a clue. Navigation is investigation."
Turn exploration into a quest — or any simulation into a hybrid LLM/deterministic CLI.
Lineage: Colossal Cave (Crowther & Woods), Scott Adams Adventures, Zork (Infocom), MUD (Bartle), LambdaMOO (Curtis).
Inherits from: simulation/ — all simulation properties plus adventure-specific state.
[!TIP] This is a general pattern. Text adventure is the reference implementation, but the same architecture powers city sims, cloud management tools, board games — anything where deterministic transforms meet creative narration.
[!TIP] Perfect for codebase archaeology. "Find where the auth bug was introduced" — that's a quest!
An adventure creates a player with state, places them in a room, and the LLM dungeon masters them around.
# player.yml
name: Alice
location: entrance-hall
inventory:
refs: # Lightweight pointers (weight: 0)
- pub/bar/brass-lantern.yml
- street/acme-catalog.yml#portable-hole
objects: # Deep copies (has weight)
- { id: notebook, name: "Notebook", weight: 0.5 }
fungibles: # Stacks
- { proto: economy/gold.yml, count: }
Inventory Protocol: See skills/inventory/ for full TAKE/DROP/BOX/BEAM operations, pointer syntax, and structural editing.
The core loop:
User: "go north"
→ DM: Updates player.location, describes the new room
User: "look around"
→ DM: Reads room YAML, narrates contents atmospherically
User: "take the rusty key"
→ DM: Moves key to player.inventory, narrates the action
The mapping:
This is Memory Palace with narrative framing and a player character.
Naturally supports multiple simultaneous participants:
# characters/
├── alice.yml # Human player 1
├── bob.yml # Human player 2
├── merchant.yml # NPC (DM-controlled)
├── guard-bot.yml # Autonomous bot (action queue)
└── oracle.yml # LLM agent with own goals
Character types:
| Type | Controlled By | Example |
|---|---|---|
| Player Character | Human via chat | Alice exploring the dungeon |
| NPC | DM (LLM) responds when addressed | Merchant sells items |
| Bot | Action queue runs autonomously | Guard patrols on schedule |
| Agent | LLM with own goals & initiative | Oracle pursues prophecies |
All coexist in the same world:
# library/ROOM.yml
occupants:
- alice # Player exploring
- bob # Another player
- librarian # NPC who answers questions
- dust-sprite # Bot that cleans autonomously
Like The Sims and Populous, you have a selection — who you're controlling right now:
selection:
mode: single # or: group, swarm
current: alice # commands go to Alice
# Or control multiple at once:
selection:
mode: group
current: [alice, bob, charlie] # "go north" moves all three
# Or a whole swarm (Populous/Dungeon Keeper style):
selection:
mode: swarm
filter: { type: imp, location: mines }
Selection commands:
| Command | Effect |
|---|---|
SELECT alice | Control Alice |
SELECT alice, bob | Control both |
SELECT ALL imps | Swarm control |
CYCLE | Next character in rotation |
Commands apply to selection:
> SELECT alice, bob, charlie
> go north
Alice goes north.
Bob goes north.
Charlie goes north.
The coherence engine orchestrates all:
graph TD
START[🎯 Quest Objective] --> R1[Enter Room]
R1 --> LOOK[👀 Look Around]
LOOK --> EXAMINE[🔍 Examine Objects]
EXAMINE --> COLLECT[📝 Collect Evidence]
COLLECT --> DECIDE{What next?}
DECIDE -->|New room| R1
DECIDE -->|Solved| END[🏆 Quest Complete]
quest/
├── ADVENTURE.yml # Quest state
├── LOG.md # Narrative journal
├── EVIDENCE/ # Collected clues
└── MAP.yml # Explored territory
adventure:
quest: "Find the authentication bug"
status: in_progress
current_room: "src/auth/"
rooms_explored: 5
clues_found: 3
hypothesis: "Session cookie not being set"
confidence: 0.7
# Adventure Log
## Day 1: Entering the Auth Dungeon
I stepped into `src/auth/` — a maze of middleware.
**Clues found:**
- `session.ts` — handles cookie creation
- `middleware.ts` — checks auth state
**Suspicion:** The cookie is created but never sent...
| Command | Action |
|---|---|
GO [direction] | Navigate |
LOOK | Describe current room |
EXAMINE [object] | Study a file |
TAKE [object] | Add to inventory |
TALK TO [npc] | Start conversation |
COLLECT [clue] | Add to evidence |
DEDUCE | Form/update hypothesis |
MAP | Show visited rooms |
INVENTORY | List held items |
DEBUG / DEBUG-ON | Enable debug mode |
DEBUG-OFF | Disable debug mode |
Toggle technical output with DEBUG-ON and DEBUG-OFF.
When debug is ON, logs include collapsible sections showing:
Example debug output:
<details>
<summary>📂 <strong>Editing CHARACTER.yml to update player location from start/ to coatroom/</strong></summary>
```yaml
# State change (CHARACTER.yml is canonical)
player:
location: start/ → coatroom/ # Character owns their location
The character file owns location state. ADVENTURE.yml mirrors it for convenience.
Files affected:
When debug is OFF, output is clean narrative without technical sections.
Customize with natural language:
> DEBUG-FORMAT Show only file operations, skip YAML, use 🔧 emoji
The format field in ADVENTURE.yml accepts natural language instructions for how to format debug output.
Trading cards can be your adventure companions:
cards_in_play:
- card: "Index Owl 🦉"
goal: "Search for cookie-related code"
- card: "Git Goblin 🧌"
goal: "Find when session handling changed"
Vision: Python CLI handles deterministic operations; LLM focuses on narrative. See README.md for full CLI vision and development plan.
| Layer | Python Does | LLM Does |
|---|---|---|
| State | Parse YAML, validate schemas | Generate content |
| Movement | Update coordinates | Narrate the journey |
| Scanning | Find pending work | Prioritize and process |
| Type | Description | Example |
|---|---|---|
| Clue | Information that might matter | "Different test runner versions" |
| Item | File worth remembering | CI config, setup.ts |
| Character | Code entity with personality | "jest.config.js — Strict about modules" |
| Map | Mental model of structure | Directory relationship diagram |
When entering any directory:
Adventures work for code exploration:
| Adventure | Investigation |
|---|---|
| Quest | Bug hunt |
| Room | Directory |
| Clue | Evidence |
| Companion | Tool card in play |
| Journal | session-log.md |
Best example: examples/adventure-4/ — The gold standard.
examples/adventure-4/pub/ — A complete social space:
pub/
├── ROOM.yml # Themeable tavern (6 themes!)
├── bartender.yml # NPC with 6 identity variants
├── pie-table.yml # Octagonal debate table
├── gong.yml # Gong of Gezelligheid
├── bar/
│ ├── bartender.yml # The omniscient bartender
│ ├── budtender-marieke.yml
│ └── cat-cave/ # TARDIS-like cat sanctuary
│ ├── ROOM.yml
│ └── 10 cats (Terpie, Stroopwafel, kittens...)
├── arcade/ # Pacman, Pong, Pinball, Fruit Machine
├── games/ # Chess, Darts, Cards
├── stage/
│ └── palm-nook/ # Multi-room character space
│ ├── study/ # Infinite typewriters, infinity desk
│ ├── gym/ # Infinite climb
│ ├── play/
│ └── rest/ # Hammock, silence cushion
└── menus/ # Drinks, snacks, buds, games
Themeable NPCs (bartender.yml):
identity:
classic_adventure:
name: Grim
appearance: "Weathered human, salt-and-pepper beard..."
space_cantina:
name: Z-4RT
appearance: "Multi-armed service droid..."
cyberpunk_bar:
name: Nyx
appearance: "Chrome-implanted bartender..."
Themeable Rooms (pub/ROOM.yml):
theme:
current: classic_adventure
themes:
classic_adventure:
name: "The Gezelligheid Grotto"
bartender: "Grim, a weathered human"
menu: ["Ale (1 gold)", "Mystery meat pie (3 gold)"]
space_cantina:
name: "The Rusty Hyperdrive"
bartender: "Z-4RT, a droid with too many arms"
menu: ["Blue milk (1 credit)", "Bantha burger"]
Rich Activities:
activities:
PERFORM: { venue: stage, effects: [tips, drinks_thrown] }
DEBATE: { venue: pie_table, rules: roberts_rules }
RING-GONG: { protocols: [once: attention, twice: emergency, thrice: mercy] }
CELEBRATE: { effects: [free_round, +morale, everyone_toasts] }
Framing Protocol (for tribute performances):
framing:
mode: [performance, celebration, tribute]
tribute_protocol:
invocation: "Before they arrive, acknowledge we're summoning them"
performance: "Depicting them as we imagine their best selves"
acknowledgment: "After they depart, note this was a tribute"
graph LR
AP[⚔️ adventure] -->|IS-A| R[🚪 room]
AP -->|companions| TC[🎴 card]
AP -->|logs to| SL[📜 session-log]
AP -->|similar to| DB[🔧 debugging]
MP[🏛️ memory-palace] -->|sibling of| AP
CLI Uplift Plan, Browser Compilation, Scott Adams History, Owl Simulation See README.md for complete development roadmap and inspiration.