MOOLLM Skills

July 8, 2026 · View on GitHub

"Skills are conventions the model follows, not code the orchestrator runs."

Userland protocols over files.

Tip

New here? Start with constructionism/ — the philosophy. Then skill/ — the meta-skill explaining how skills work. Then explore room/ — it has everything!

INDEX.md — Tour of all 128 skills and how they connect. INDEX.yml — Machine-readable registry.

Eight Architectural Extensions

MOOLLM extends Anthropic's skill model with eight innovations:

#ExtensionWhat It AddsProof
1InstantiationSkills as prototypes creating instances. Not just prompts — living programs.adventure/ → adventure-4/ with 150+ files
2Multi-Tier PersistencePlatform (ephemeral) → Narrative (append) → State (edit) → MOO-Maps (GLANCE/CARD/SKILL/README/examples/templates/source).6000+ line session logs, persistent room state
3K-linesNames as semantic activation vectors (Minsky)."Palm" activates entire soul, history, relationships
4Empathic TemplatesSmart generation, not string substitution.Biscuit's description generated from traits
5Speed of LightMany turns in one call, minimal tokenization.33-turn Fluxx, 21-turn cat prowl
6CARD.ymlMachine-readable interface with advertisements.Every skill exposes methods, tools, state schema — see card/
7Ethical FramingRoom-based inheritance of performance context.pub/stage/ inherits framing: performance
8Ambient SkillsAlways-on behavioral shaping (NO-AI hygiene suite).no-ai-slop, postel, robust-first — see skill/

The key insight: Skills aren't documentation. They're programs. The LLM is eval().

Full explanation: MOOLLM Eval Incarnate Framework

Design by Accretion

This library is not specified top-down; it accretes — the method Maxis designer Chris Trottier named in describing how The Sims came together: layer, accumulate, let it grow, tune late. You can watch the signature in this very directory: the skill count in this README, in INDEX.yml, and on disk usually disagree, because the pile grows faster than its catalogs. That lag is the method working, not an error — the tuning pass (recount, re-index, prune, compose) comes after the layers.

The vocabulary of accretion is itself accreted here as skills: cauldron/ (melt → stir → ladle), play-learn-lift/ (explore → stabilize → publish), simulator-effect/ (implication over simulation — Will Wright). The Repo Show planted a flagship episode on exactly this, with Chris as the guest who lived it: Design by Accretion — including a question from Palm, a character who lives inside this accreting design: from inside, how do you tell a heap from an organism?

Skill Index (128 skills)

Philosophy & Core Concepts

SkillOne-liner
moollm/The soul of MOOLLM. Self-explanation, help, navigation
leela-ai/Manufacturing Intelligence. Leela AI develops MOOLLM for practical devops and exploring industrial applications
manufacturing-intelligence/The slogan unpacked -- seven readings of one phrase
skill/The meta-skill. How skills work, evolve, compose
k-lines/Minsky's K-lines — names that activate conceptual clusters
constructionism/Learn by building inspectable things (Papert, Kay, Logo)
prototype/Self-language philosophy — clone, don't instantiate
yaml-jazz/Semantic YAML where comments are data
postel/The Robustness Principle — be liberal in acceptance
robust-first/Survive first, be correct later (Dave Ackley)
coherence-engine/LLM as consistency maintainer & orchestrator
speed-of-light/Many turns in one call — instant telepathy
simulator-effect/Implication beats simulation (Will Wright) — imagination renders
society-of-mind/Intelligence emerges from many simple agents (Minsky)
procedural-rhetoric/Rules persuade, structure IS argument (Bogost)
schema-mechanism/Drescher's causal learning extended with LLM semantics

Formats & Structure

SkillOne-liner
plain-text/Text files are forever — no lock-in, no corruption
markdown/Readable raw AND rendered
format-design/How to design formats that succeed
sniffable-python/Structure Python so LLMs understand in 50 lines
naming/Big-endian file naming as semantic binding
empathic-expressions/Intent-based code interpretation
empathic-templates/Smart templates with semantic understanding

Methodology (How to Work)

SkillOne-liner
bootstrap/Wake up, orient, warm the context
play-learn-lift/Explore → Learn → Share wisdom
planning/Flexible task decomposition
plan-then-execute/Frozen plans with human approval gates
sister-script/Document-first automation
research-notebook/Structured investigation with sources
debugging/Hypothesis-driven bug hunting
code-review/Systematic code analysis

Spatial (Room/Card System)

SkillOne-liner
room/Directories as activation contexts
card/Capabilities as portable, playable cards
container/Intermediate scopes — inheritance without navigation
logistic-container/Factorio-style logistics boxes
adventure/Narrative room exploration
memory-palace/Spatial knowledge organization
data-flow/Rooms as pipeline nodes (THROW/INBOX)
return-stack/Navigation history as continuation
multi-presence/Same card active in multiple rooms
exit/Connections between rooms
object/Things in the world
context/Activation environment

Characters & Identity

SkillOne-liner
character/Core patterns — home, location, relationships
incarnation/Gold-standard creation — characters write their own souls
persona/Identity layers (WHO vs WHAT)
soul-chat/Everything speaks — multi-voice dialogues
mind-mirror/Personality via four Thought Planes (Leary)
hero-story/Safe K-line references to real people
representation-ethics/Ethics of simulating people
visualizer/Semantic image generation with metadata

Animal Characters

SkillOne-liner
cat/Feline behavior — trust earned, charms, the forbidden belly
dog/Canine behavior — loyalty, pack dynamics, unconditional love

Role Skills (Professions)

SkillOne-liner
bartender/Pour drinks, manage tabs, know everyone's secrets
budtender/Cannabis-specialized — strains, terpenes, responsible service

Game Mechanics (Sims-style)

SkillOne-liner
simulation/Central hub — turns, party, selection, world
time/Simulation turns vs LLM iterations
buff/Temporary effects (curses = negative buffs)
needs/Dynamic motivations — hunger, energy, fun
party/Companions and group dynamics
action-queue/Sims-style task queue
advertisement/Objects announce what they can do
world-generation/Questions create places

Economy & Scoring

SkillOne-liner
economy/Currency, trade, gold flow
probability/Success calculation from stats
scoring/Achievement valuation
reward/Dynamic achievement rewards

Decision & Deliberation (Mike Gallaher's Methodology)

SkillOne-liner
adversarial-committee/Committee of opposing personas force genuine debate
debate/Structured multi-perspective deliberation
roberts-rules/Parliamentary procedure prevents short-circuiting
rubric/Measurable criteria for scoring decisions
evaluator/Independent assessment without debate context

"Everything is a story. No single story is true — but the ensemble approximates actionable wisdom."

See: designs/mike-gallaher-ideas.md

Memory & Context

SkillOne-liner
summarize/Compress without losing truth
honest-forget/Graceful memory decay
session-log/Human-readable audit trail
scratchpad/Working memory

Communication

SkillOne-liner
postal/Intra-world mail system — letters, packages, inboxes, THROW

System & Recovery

SkillOne-liner
self-repair/Checklist-based healing
storytelling-tools/Narrative capture — notebooks, letters, photos
runtime/Execution environment
image-mining/Extract resources from images

Goals & Subjective

SkillOne-liner
goal/Objectives and quests
subjective/First-person experience

Skill Tiers

TierTools RequiredExamples
0None (pure prompt)Writing styles, reasoning patterns
1File read/writeMost skills
2+ TerminalDebugging, code-review

Principle: Use the lowest tier possible.

Skill Anatomy (Required Structure)

skills/
  my-skill/
    README.md         # Human entry point (GitHub renders this)
    SKILL.md          # Full spec with YAML frontmatter
    CARD.yml          # Machine-readable interface definition
    *.tmpl            # Optional: templates at root level

Every skill has three required files:

FilePurpose
README.mdQuick overview, links (GitHub renders this)
SKILL.mdFull protocol with YAML frontmatter (name, tier, allowed-tools)
CARD.ymlInterface definition: methods, tools, state, advertisements

Skills may be published independently (zip or bare SKILL.md). Publishing policy: designs/SKILL-PUBLISHING-POLICY.md — self-contained SKILL.md, related skills documented, standard "Part of MOOLLM" blurb, metadata compatible with Anthropic/Agent Skills.

Quick Start

Use a Skill

  1. Read the skill's README.md
  2. Read SKILL.md for full protocol
  3. Copy *.tmpl files to your working directory
  4. Follow the documented protocol

Create a New Skill

  1. Create skills/my-skill/ directory
  2. Copy templates from skills/skill/:
    • README.md.tmpl → README.md
    • SKILL.md.tmpl → SKILL.md
  3. Create CARD.yml with methods, tools, state
  4. Fill in template variables
  5. Register in INDEX.yml

Or just tell the LLM: "Create a new skill called 'my-skill' using the skill skill."

Meta Files

FilePurpose
INDEX.ymlMachine-readable skill registry
ROOM.ymlThe Skill Nexus — this directory as a metaphysical room
skill/The meta-skill with templates and protocols

The Skill Nexus as Shared Space

This skills/ directory is a shared room accessible from any adventure. Unlike adventure-specific rooms, it lives at the repo root and connects to ALL adventures.

Path Variables

Instead of counting ../../../ levels, use path variables that resolve at runtime. See kernel/NAMING.yml for the full specification.

VariableResolves ToUse Case
REPO-SCOPEDShared across all adventures
$REPO/moollm/Repository root
$SKILLS/moollm/skills/Skill definitions
$KERNEL/moollm/kernel/Core protocols
$DESIGNS/moollm/designs/Architecture docs
ADVENTURE-SCOPEDPlugged in at runtime
$ADVENTURE/Current adventureFrom startup.yml
$CHARACTERS/$ADVENTURE/characters/Character alcoves
$ANIMALS/$CHARACTERS/animals/Animal sanctuary
$PERSONAS/$ADVENTURE/personas/Mask wardrobe
$PUB/$ADVENTURE/pub/The gathering place
$COATROOM/$ADVENTURE/coatroom/Transformation room
$START/$ADVENTURE/start/Origin point

Dynamic Binding: Skills use $ANIMALS/dog-biscuit/ — adventure provides the concrete path. Two-way links dovetail: skill→adventure and adventure→skill resolve consistently.

Example Usage

# In skills/ROOM.yml — back-links to adventure
exits:
  coatroom:
    destination: $COATROOM/
    
relationships:
  maurice:
    location: $COATROOM/mannequin.yml
    
# In adventure YAML — forward-links to skills
exits:
  east:
    destination: $SKILLS/

Concrete Resolution (when adventure-4 is active)

Path VariableResolves To
$COATROOM/mirror.ymlexamples/adventure-4/coatroom/mirror.yml
$CHARACTERS/animals/monkey-palm/examples/adventure-4/characters/animals/monkey-palm/
$SKILLS/character/skills/character/
DirectionDestination
UpProject Root
Siblingkernel/ — Low-level protocols
Siblingschemas/ — Data formats
Siblingdesigns/ — Historical archives

The Intertwingularity

"Everything is deeply intertwingled." — Ted Nelson

graph TB
    subgraph FOUNDATION["FOUNDATION"]
        MOOLLM[moollm<br/>self + help]
        SKILL[skill<br/>meta-skill + 8 extensions]
        KLINES[k-lines<br/>semantic activation]
        BOOT[bootstrap<br/>compile + init]
        MOOLLM --> SKILL
        SKILL --> KLINES
        MOOLLM --> BOOT
    end

    subgraph FORMAT["FORMAT"]
        YAML[yaml-jazz<br/>comments = data]
        POSTEL[postel<br/>liberal in, clean out]
        NAMING[naming<br/>big-endian K-lines]
        YAML --> POSTEL
        YAML --> NAMING
    end

    subgraph AMBIENT["AMBIENT SKILLS — always on"]
        SLOP[no-ai-slop]
        GLOSS[no-ai-gloss]
        SYCO[no-ai-sycophancy]
        HEDGE[no-ai-hedging]
        MORAL[no-ai-moralizing]
        ROBUST[robust-first]
    end

    subgraph SPACE["SPATIAL"]
        ROOM[room<br/>directory = context]
        ADV[adventure<br/>exploration]
        MP[memory-palace<br/>method of loci]
        OBJ[object<br/>interactable atoms]
        ROOM --> ADV
        ROOM --> MP
        ROOM -->|"contains"| OBJ
    end

    subgraph CHAR["CHARACTERS"]
        CHARACTER[character<br/>body + home]
        PERSONA[persona<br/>masks]
        NEEDS[needs<br/>Sims motives]
        MM[mind-mirror<br/>Leary + Sims]
        INCARNATION[incarnation<br/>creation protocol]
        CHARACTER --> PERSONA
        CHARACTER --> NEEDS
        CHARACTER --> MM
        INCARNATION -->|"creates"| CHARACTER
    end

    subgraph SIMS["SIMS PIPELINE"]
        ADS[advertisement<br/>objects broadcast actions]
        AQ[action-queue<br/>task scheduler]
        ECON[economy<br/>MOOLAH + karma]
        ADS --> AQ
        AQ --> ECON
    end

    subgraph SIM["SIMULATION"]
        SIMULATION[simulation<br/>turns + state]
        TIME[time<br/>turns ≠ iterations]
        EXP[experiment<br/>SIMULATE/EVALUATE]
        MICRO[micropolis<br/>city simulation]
        SOL[speed-of-light<br/>many turns, one call]
        SIMULATION --> TIME
        SIMULATION --> EXP
        SIMULATION --> MICRO
        SOL -->|"accelerates"| SIMULATION
    end

    subgraph DELIBERATION["DELIBERATION"]
        AC[adversarial-committee<br/>forced debate]
        DEBATE[debate<br/>structured]
        EVAL[evaluator<br/>independent]
        RUBRIC[rubric<br/>criteria]
        AC --> DEBATE
        EVAL --> RUBRIC
    end

    subgraph INTROSPECTION["INTROSPECTION"]
        MIRROR[cursor-mirror<br/>watch yourself think]
        SNITCH[skill-snitch<br/>security audit]
        SCHEMA[schema-mechanism<br/>causal learning]
        MIRROR --> SNITCH
    end

    subgraph METHOD["METHODOLOGY"]
        PLL[play-learn-lift<br/>explore, pattern, share]
        SISTER[sister-script<br/>doc = automation]
        LOG[session-log<br/>audit trail]
    end

    %% Cross-cluster: how everything connects
    FOUNDATION -->|"defines"| FORMAT
    FOUNDATION -->|"compiles"| AMBIENT
    FORMAT -->|"structures"| SPACE
    CHARACTER -->|"lives in"| ROOM
    CHARACTER -->|"advertises via"| ADS
    SIMULATION -->|"manages"| CHARACTER
    ADV -->|"runs on"| SIMULATION
    MICRO -->|"inherits"| ADV
    AC -->|"debates about"| SIMULATION
    SCHEMA -->|"learns from"| SIMULATION
    EXP -->|"evaluated by"| EVAL
    PLL -->|"records to"| LOG
    MIRROR -->|"observes"| SIMULATION
    AMBIENT -.->|"shapes all output"| DELIBERATION
    AMBIENT -.->|"shapes all output"| SIM

Every skill connects to others. The foundation defines how skills work. Format shapes how data flows. Ambient skills shape every output. Characters live in rooms, advertise actions, and are managed by simulations. Deliberation debates simulation outcomes. Introspection watches it all. Methodology records what happens. Navigate freely.

📚 See Also

Protocols & Symbols

Kernel (Infrastructure)

Schemas (Shapes)