The grammar rules that make MOOLLM's file system object-oriented. Plural directory names declare element type; UPPERCASE marker files declare interface exports (COM-style, minus the UUIDs); directories are implementation classes exporting every interface whose marker file sits at their root.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
A direct command skips the review prompt. Inspect the source before running it.
The grammar rules that make MOOLLM's file system object-oriented. Plural directory names declare element type; UPPERCASE marker files declare interface exports (COM-style, minus the UUIDs); directories are implementation classes exporting every interface whose marker file sits at their root.
["Smalltalk (Alan Kay et al., Xerox PARC, 1970s) โ classes, messages, image-as-object","Self (Ungar & Smith, 1987) โ prototype delegation; objects without classes","NeWS (James Gosling & David S. H. Rosenthal, Sun, 1986) โ PostScript window system with prototype delegation in production","TNT: The NeWS Toolkit (Owen Densmore & David S. H. Rosenthal, Sun) โ class hierarchy built on NeWS's prototype foundation; the patent and the conceptual ancestor","COM / XPCOM โ multiple interfaces per class; QueryInterface","Ted Nelson โ intertwingled systems; every node can link anywhere","MOOLLM โ the actually-existing MOOLLM file system, already using this grammar before it was named"]
File System as Object โ Directories, Interfaces, and the UPPERCASE Marker Grammar
"A directory named <things>/ is saying: the things in here are of the type <thing>. A file named FOO.md at a directory's root is saying: this directory exports the FOO interface. The directory IS an implementation class; the UPPERCASE marker files ARE its interface declarations; no UUIDs needed because the filename IS the class name."
This skill formalizes a grammar that already exists throughout MOOLLM. Every skill you've read โ skill, prototype, card, schema, adventure, room, character, incarnation โ uses these rules implicitly. The other meta-skills USE the grammar; this skill NAMES it so newcomers (human or LLM) can read MOOLLM directory trees fluently without pattern-matching each one from scratch.
The Three Rules
Rule 1: Plural-named directories declare element type
A directory named <things>/ declares "the things in here are of type <thing>."
skills/ โ each child is (presumably) a skill
biomes/ โ each child is a biome
mechanisms/ โ each child is a schema mechanism
recipes/ โ each file is a recipe
gotchas/ โ each file is a gotcha
runbooks/ โ each file is a runbook
instances/ โ each child is an instance
stacks/ โ each child is a stack
protocols/ โ each file is a protocol
rooms/ โ each child is a room
characters/ โ each child is a character
"Presumably" is doing real work here: the plural name is a declaration of intent. To confirm a specific child actually IS of that type, look for its interface marker file (see Rule 3). Auxiliary content (templates, indexes, READMEs) can also live in a plural container without being instances โ their absence of a marker file is the tell.
Rule 2: Singular-named directories are instances
A directory named <thing>/ (singular, kebab-case or single word) is an instance. The directory name is the instance's identifier.
skills/skill/ โ instance named "skill" (a specific skill, happens to BE the meta-skill about skills)
skills/biome/ โ instance named "biome"
skills/gcs/ โ instance named "gcs"
mechanisms/json-schema/ โ instance named "json-schema"
instances/leela-zion2-dev-0/ โ instance named "leela-zion2-dev-0"
A file named FOO.md or FOO.yml (UPPERCASE filename) at a directory's root declares that the directory EXPORTS the FOO interface.
skills/gcs/
โโโ SKILL.md โ "I export the SKILL interface"
โโโ CARD.yml โ "I export the CARD interface"
โโโ GLANCE.yml โ "I export the GLANCE interface"
โโโ README.md โ "I export the README interface (human landing)"
โโโ recipes/ โ plural: "I contain recipes"
โโโ gotchas/ โ plural: "I contain gotchas"
โโโ ...
The UPPERCASE convention visually distinguishes interface declarations from ordinary content. The presence of the file IS the declaration โ no separate registry required.
One directory exports many interfaces. Just like a COM class implements many interfaces, a MOOLLM skill directory typically exports SKILL + CARD + GLANCE + README + sometimes CHARACTER + EXPORTS + PROTOTYPES โ all simultaneously, via multiple marker files.
The COM Analogy
This is Henry's framing, which anchors the whole grammar:
COM concept
File-system-object equivalent
Implementation class (C++ / IDL class)
Directory โ the <thing>/ singular name
Interface declaration (IDL interface IFoo)
UPPERCASE marker file โ FOO.md or FOO.yml
Interface UUID (IID_IFoo = {01234567-89ab-...})
Not needed โ the filename IS the identifier
Multiple interfaces per class
Multiple UPPERCASE marker files per directory
QueryInterface(IID_IFoo, &ptr)
cat <dir>/FOO.md (or "does this dir export FOO?" = test -e <dir>/FOO.md)
Class factory (CoCreateInstance)
Prototype instantiation (see skill-instantiation-protocol.md, delegation-object-protocol.md)
Aggregation (one class delegating to another)
DOP prototype chain via inherits: or PROTOTYPES.yml
COM's big wins were: (a) multiple interfaces per implementation, (b) language-neutral contracts, (c) strong versioning via UUIDs. MOOLLM keeps (a) and (b), swaps (c) for readable UPPERCASE names because we want humans and LLMs to read the tree, not a binary-stamped registry.
COM's big losses were: opaque UUIDs, fiddly registration, reliance on a global registry, Windows-centric tooling. MOOLLM avoids all four by putting the declarations in the filesystem itself.
Why UPPERCASE
The UPPERCASE-filename convention (SKILL.md, CARD.yml, MECHANISM.yml, ALERT.yml, PROTOTYPES.yml) does several things at once:
Visual grep. In any ls output, UPPERCASE files leap out. Structural signposts pop; ordinary content recedes.
Class-vs-content distinction. Lowercase filenames (camera-configs.json, my-recipe.md) are content; UPPERCASE is meta โ "I am telling you what this directory IS."
Namespace cleanliness.foo.md (lowercase) is a note about foo; FOO.md (uppercase) is a declaration that this directory implements FOO. They don't collide, and their roles are clear at a glance.
Future-proof tooling. A validator script can simply glob for [A-Z]*.{md,yml} at a directory root to find all exported interfaces. No manifest file, no parsing.
Big-endian naming. The UPPERCASE prefix groups structural files together alphabetically at the top of ls output.
When NOT to use uppercase. Normal content, per-instance data (state.yml, history.md), templates (RUN.yml.tmpl โ the .tmpl is the tell that it's a template, not a declaration), ordinary documentation (troubleshooting.md).
Interface marker files come in three flavors, each picked on purpose:
Extension
Use when
Why
.yml
The interface is data-shaped (fields, structure, lists). Default choice.
Comments are first-class and encouraged (see yaml-jazz). The same file can carry structure, semantic hints, lineage notes, and authorial voice simultaneously.
.md
The interface is narrative (explains, teaches, guides).
Markdown supports prose + structure + code blocks. SKILL.md, README.md are the canonical narrative interfaces.
.json
A downstream tool needs it (jq pipelines, ajv validators, frontend code reading an API) and no human is expected to author it by hand.
Universal tool support. Smaller parser footprint. But: no comments โ strips human intent.
The rule: YAML is preferred, JSON is welcome, comments are encouraged.
When to prefer which
Authoring by hand? โ YAML. Every time.
Tool reads it, humans rarely? โ JSON is fine.
Both? โ YAML is canonical; emit JSON on a hook if needed.
Narrative / teaching / guiding? โ Markdown.
Why YAML over JSON for MOOLLM interface declarations
YAML gives us three-audiences support (humans, LLMs, machines โ see the yaml-jazz skill):
# CARD.yml โ this is a comment โ it carries semantic information for# humans AND LLMs, even though machines technically ignore it.card:id:exampletagline:"The tagline is the elevator pitch."# โ Next audience note: rule-of-thumb is one sentence, present tense.
Strip the comments and you have the same data. Keep them and you have intent alongside the data โ why the fields are there, what alternatives we considered, which examples are load-bearing. That's the yaml-jazz difference, and it's why every UPPERCASE marker file in MOOLLM defaults to YAML.
.json next to .yml canonical, regenerated on change.
Template interfaces
CARD.yml.tmpl, SKILL.md.tmpl
.tmpl suffix signals placeholder content; not a real declaration.
If you're introducing a new UPPERCASE interface type and you have to choose: FOO.yml is the default. Reach for .md if the content is prose-heavy; reach for .json only when a tool demands it.
The yaml-jazz skill is a hard dependency
This skill inherits conceptually from yaml-jazz: "Semantic YAML โ comments as data; three audiences." The two skills together give MOOLLM its authoring grammar:
file-system-object says: directories are classes, UPPERCASE files are interfaces.
yaml-jazz says: inside those interface files, comments are first-class semantic carriers.
Together they turn an ls + cat session into a legitimate class browsing experience.
Nesting โ Sub-skills and Categorization
Following the grammar, a directory can nest further plurals inside itself:
skills/biome/ โ singular (instance of skill type)
โโโ SKILL.md โ exports SKILL interface
โโโ CARD.yml โ exports CARD interface
โโโ biomes/ โ plural: "this skill contains biomes"
โ โโโ registry.yml โ (not UPPERCASE โ data, not interface)
โ โโโ gateways.yml โ (not UPPERCASE โ data, not interface)
โ โโโ convention.yml โ (not UPPERCASE โ data, not interface)
โ โโโ families.yml โ (not UPPERCASE โ data, not interface)
โโโ templates/
โโโ BIOME/ โ the template IS named UPPERCASE because it's a TYPE template
โโโ SKILL.md.tmpl
โโโ CARD.yml.tmpl
โโโ ...
A skill can also contain sub-skills:
skills/parent-skill/
โโโ SKILL.md
โโโ CARD.yml
โโโ skills/ โ plural: sub-skills live here
โโโ child-skill/
โโโ SKILL.md
โโโ CARD.yml
And so on recursively. The grammar holds at any depth.
Skills as categories
When a skill contains a skills/ subdirectory, the parent skill plays two roles at once:
It's still a skill itself (via its own SKILL.md / CARD.yml).
It's also a CATEGORY for the sub-skills โ the parent directory name is the category name; its child skills belong to that category by virtue of their path.
skills/gardening/ โ the gardening skill AND the gardening category
โโโ SKILL.md โ the parent skill's full protocol
โโโ CARD.yml
โโโ skills/ โ sub-skills (i.e., members of the gardening category)
โโโ composting/
โ โโโ SKILL.md
โ โโโ CARD.yml
โโโ pruning/
โ โโโ SKILL.md
โ โโโ CARD.yml
โโโ seed-starting/
โโโ SKILL.md
No separate "category" concept is needed โ the parent skill IS the category. A sub-skill participates in the category simply by living under skills/<parent>/skills/<child>/. Cross-category sub-skills (a skill that belongs to two categories) are handled by DOP inheritance, not by duplicating directories: child-skill lives in one canonical path, and other categories reference it via inherits: chains.
When to nest vs flatten. Put a skill under a parent when:
It's only meaningful inside the parent's context (e.g., biome's biomes/ stubs โ each stub is only a reference biome for this mother skill).
It's a sub-concern of the parent that isn't reused elsewhere yet.
Pull it up to a top-level skill (lift it) when:
It becomes reusable across unrelated contexts.
Other skills start wanting to inherit from it.
Its audience broadens beyond users of the parent skill.
This is the same PLAY-LEARN-LIFT arc applied to skill hierarchy: sub-skills start nested (low-cost), get promoted when they earn independence.
Skill Lookup and Inheritance Through Tree Containment
When one skill references another โ via inherits:, related:, or a prose mention โ the reference can take two forms:
Form
Example
When to use
Bare name
inherits: [biome]
Most authoring. Short, Postel-friendly, resolver-driven.
Unambiguous path fragment
see: moollm/skills/biome/SKILL.md
When you need to disambiguate, cross-repo-reference, or help a reader find the exact file.
Both are valid. MOOLLM's resolver is expected to accept either and find the right skill, following Postel's robustness principle: be liberal in what you accept, conservative in what you emit. When authoring by hand, reach for the unambiguous form when you easily can; reach for the bare name when the context makes it obvious.
What counts as a skill โ duck typing on marker files
A directory participates in the MOOLLM object system to the extent its UPPERCASE marker files say it does. There are three distinct levels of participation, and a directory can sit at any of them:
Level
Minimum markers
What works
What doesn't
๐ Full skill
SKILL.md + CARD.yml (+ usually GLANCE.yml, README.md)
Everything: dispatch, narrative, semantic-image-pyramid reading, ambient inheritance when contained
โ
๐ Dispatchable object
CARD.yml alone
Dispatch: methods, advertisements, k-lines, state, navigation all work. queryInterface(dir, 'card') returns the sniffable interface. Inheritance declared in CARD.yml's inherits: chain resolves normally.
No narrative/protocol doc; no SKILL.md-level reading-order entry point (GLANCE โ CARD โ SKILL โ README collapses to GLANCE โ CARD)
๐ฆ Scope-declared
skills/<name>/ container (or biomes/<name>/, characters/<name>/, etc.) declares the TYPE of the directory
Type-level discovery works (enumeration recognizes it as of the plural's type); becomes dispatchable the moment a CARD.yml appears
Without CARD or SKILL, there's nothing to dispatch yet โ it's a stub, not an error
A CARD-only directory is a first-class object. It isn't a "degraded skill" โ it's a lightweight form for when narrative isn't warranted. A directory with just CARD.yml can still:
Be invoked via its methods: block
Advertise capabilities via advertisements:
Activate k-lines via k_lines:
Declare inheritance via inherits: (resolver walks the chain the same way)
Participate in DOP delegation
Be the target of related: from other skills
skills/my-thing/ โ dispatches just fine
โโโ CARD.yml โ this alone carries methods, ads, k-lines, navigation
โ queryInterface(., 'card') โ CARD.yml
โ queryInterface(., 'skill') โ null (no SKILL.md โ that's OK)
This matters because it removes a friction: you don't need to write a 300-line SKILL.md just to make something dispatchable. Write the CARD when you have methods to declare; write the SKILL when you have a protocol to teach. The two files answer different questions ("what can this DO?" vs "how does this WORK?"), and not every object needs a teaching narrative.
๐ฆ Quack-quack duck typing โ a directory is an object to the degree its marker files quack. SKILL.md is the full orchestra; CARD.yml is a solid string quartet; skills/<dir>/ with no markers is a stage with a name on it. All three are legitimate; each does a different amount of work.
Any additional UPPERCASE marker file adds another interface: CHARACTER.yml makes it incarnation-ready; MECHANISM.yml registers it with schemapedia; ALERT.yml makes it a branch-as-object alert. The directory exports all of them simultaneously, in classic COM-style multiple-interface fashion.
๐ด The CARD is a rich pun.CARD.yml names a surface that is โ simultaneously โ an interface definition (IDL / OLE Control / ActiveX TypeLib), a HyperCard (Atkinson-style navigable unit with fields/buttons/scripts), an actor (Hewitt-style message-receiver with local state), a thread (a dispatchable unit of control), a message dispatching surface (Smalltalk/Self-style), and a portable token (trading card + business card). The richness is load-bearing; see skills/card/SKILL.md ยง "The Card Pun Stack" for the full lineage.
The lookup scope โ two searches, one upward walk
A bare reference like inherits: [biome] is resolved by walking outward from the referring skill's location. The walk performs two independent searches at every level. Both use the same parent-walking loop; they return different things.
referring_skill_dir/
โ up one level
referring_skill_dir's parent/
โ up one level
...
At each step, check two things:
Search 1 โ scoped skills/ directory at this level
If the current directory contains a skills/ subdirectory, it declares scoped sub-skills โ skills visible by bare name from anywhere under this directory. The containing directory does not itself have to be a skill โ any directory can host a scoped skills/ subdirectory (a project root, a repo root, an adventure, a biome, a runbook bundle, anything).
some-project/ โ not a skill (no SKILL.md), but...
โโโ data/
โโโ README.md
โโโ skills/ โ scopes skills to this project
โโโ biome/ โ resolves when anything under some-project/ says `biome`
โโโ ingest/
This is "I bring my own skill library" โ a directory can define its own private skills without publishing them, and things inside it pick them up automatically. The same pattern at the repo root (<repo>/skills/), the MOOLLM root (moollm/skills/), and even sibling repos mounted nearby (~/.../Leela/git/*/skills/) all follow the same rule โ skills/ is a scoping container wherever it appears.
Search 2 โ this directory IS an ambient parent skill
If the current directory has a SKILL.md (or other UPPERCASE marker file) at its root, it is itself a skill, and its conventions become ambient for anything inside it. A skills/ subdirectory is not required โ any skill directory can contain arbitrary nested content, and that content inherits the parent's conventions.
moollm/skills/biome/ โ SKILL.md present โ this dir is a skill
โโโ SKILL.md
โโโ CARD.yml
โโโ biomes/
โโโ gcp/ โ inside `biome`'s tree โ biome is an ambient parent;
โโโ GLANCE.yml โ gcp's conventions fall back to biome's conventions
gcp doesn't need to declare inherits: biome for ambient containment inheritance โ being located under moollm/skills/biome/ is already the signal. (Explicit inherits: biome is still encouraged when authoring for clarity.)
Putting the two searches together
# A directory "quacks like a skill" if it has any recognized marker:
is_skill_like(d) := exists(d / "SKILL.md") or exists(d / "CARD.yml")
# (more broadly: any UPPERCASE.{md,yml} at root makes d a
# dispatchable object of that interface's type โ this is
# the general rule; SKILL/CARD are the common skill pair.)
resolve_skill(name, from_dir):
# Phase 1 โ local tree walk (Search 1 + Search 2 at each step)
d = from_dir
while d is not filesystem-root:
# Search 1: scoped `skills/` subdir at this level
if is_skill_like(d / "skills" / name):
return d / "skills" / name
# Search 2 / self-match: this dir IS the skill we're looking for
if basename(d) == name and is_skill_like(d):
return d
d = parent(d)
# Phase 2 โ mounted-repo search (Cursor workspace roots, etc.)
for repo in mounted_workspace_roots():
if is_skill_like(repo / "skills" / name):
return repo / "skills" / name
return NOT_FOUND
ambient_parents(dir):
# Independent of resolve โ these are conventions inherited by tree containment.
# Any ancestor that quacks like a skill counts as an ambient parent.
d = parent(dir)
while d is not filesystem-root:
if is_skill_like(d):
yield d # d is an ambient parent skill (SKILL.md OR CARD.yml at root)
d = parent(d)
The two searches compose: resolve_skill finds the skill a name points at; ambient_parents yields the chain of enclosing skills whose conventions apply regardless of any explicit declaration.
Mounted workspace roots โ the distributed skill path
When MOOLLM runs under Cursor (or any editor with multiple mounted workspace roots), the set of mounted repos becomes the final search path for skill resolution. Each mounted repo contributes its skills/ subdirectory to the resolver โ treat the whole set as one flat union, walked in a stable order after the local-tree walk fails.
Typical sources that get mounted and contribute skills:
Category
Example
Shape
MOOLLM core
moollm/skills/
The baseline skill library (biome, schema, card, skill, yaml-jazz, โฆ)
Customer-specific
central/skills/, <customer-repo>/skills/
Private per-customer overlays that inherit from MOOLLM mothers
Distributed / team
<org-infra>/skills/, <team-shared>/skills/
Organization-wide libraries shared across projects
A developer's own scratch library of work-in-progress skills
Per-project
<some-project>/skills/
Skills scoped to just this project (same shape as any Search-1 hit)
The resolver treats them all as equal-citizen skill roots. There's no distinction in the grammar between "official MOOLLM" and "third-party" โ they're all just <some-repo>/skills/<some-name>/. The only difference is what the user mounted; the grammar doesn't care.
Order and shadowing across repos
Within Phase 2, the order is:
The repo containing the referring skill (if any) โ checked first, because "my own repo" is closer.
Other mounted repos, in a stable order (e.g., alphabetical, or Cursor's workspace order, or explicitly configured).
MOOLLM core last โ it's the fallback, the most general library.
Closer still shadows farther. If your own repo's skills/biome/ exists, it wins over moollm/skills/biome/. This is how overlays work: a customer repo can locally customize a MOOLLM skill just by defining a skill of the same name in its own skills/ directory, with inherits: [moollm/skills/biome] if it wants to extend rather than replace.
Cursor-specific behavior
Cursor's workspace model is ideal for this pattern. The user mounts central/, moollm/, autotest/, leela-alerts/, and some customer-specific repo; MOOLLM running as an agent sees all of them at once and can resolve skills across the entire mounted set. The user didn't have to publish anything anywhere โ mounting the repo made its skills findable.
Graceful degradation when a repo isn't mounted
If a skill name resolves to <some-repo>/skills/<name>/ on developer A's machine (because they have the repo mounted) but not on developer B's (who doesn't), MOOLLM should:
Treat the bare name as unresolved when the repo is missing.
Fall through to whatever else is reachable.
Emit a one-line warning identifying the expected repo, so the developer can mount it if needed.
This is the robust-first principle โ not crashing on a missing repo; degrading gracefully with a clear signpost.
Distribution as a cooperative ecosystem
The practical effect is that MOOLLM skills form a distributed ecosystem by default. Publish a skill repo; mount it; use it. No registry, no package manager, no version-coordination protocol at the skill layer. Git is the package manager; the filesystem is the registry; skills/<name>/ is the canonical location; bare-name lookup is the import.
Versioning, when needed, is handled at the git-commit/tag/branch level โ mount a specific ref of the skill repo to pin to a specific version. For most use cases, "mount current main" is fine and mirrors how MOOLLM already operates.
Parent-dir containment = relatedness signal
A skill that lives in a parent directory of another skill is, by convention, related to it. Containment in the filesystem tree is a meaningful signal, not an accident of organization:
moollm/skills/biome/biomes/gcp/ โ the gcp biome is a sub-skill of biome; it inherits biome's conventions because it sits under biome/biomes/ (plural container: biomes/, plural of its parent's element type).
moollm/skills/schema/schemas/mechanisms/json-schema/ โ the json-schema mechanism belongs to the schema meta-skill by virtue of its path.
central/skills/gcs/ next to moollm/skills/biome/ (sibling repos on the same developer's filesystem) โ gcs can reference biome by name because MOOLLM searches outward; biome is found in the sibling-repo fallback step.
Inheritance through tree containment is a real mechanism, not just a suggestion. A skill at A/B/C/foo/ inherits the ambient behaviors and conventions of any of A/B/C/, A/B/, A/ that are themselves skills (have SKILL.md at their root) โ in that order โ before falling through to repo-wide / MOOLLM-wide defaults. This parallels:
Python โ module resolution via sys.path, innermost __init__.py wins.
JavaScript / Node.js โ node_modules walked up from the current file toward the repo root.
CSS โ cascade from parent selector to child; inner rules override outer.
DOM โ event bubbling; the target's ancestors hear the event unless stopped.
Lexical scope in Self / Scheme / Lisp โ name lookup walks the enclosing environment chain.
.gitignore โ patterns in a nested dir compose with (and override) ancestors'.
COM's aggregation โ an outer object delegates QueryInterface to an inner one until one responds.
MOOLLM's skill resolver is the filesystem analog of all of these at once.
Postel's law, applied
When emitting skill references, prefer the unambiguous path fragment:
When accepting skill references (writing a resolver, reading foreign skills), accept all of:
inherits: [biome] # bare nameinherits: [moollm/skills/biome] # unambiguous fragmentinherits: [../biome/] # relative path (works if both are siblings)inherits: [./skills/biome] # workspace-relativerelated:biome# scalar, not listrelated:"biome, card, schema"# comma-separated string
The resolver normalizes all of these to the same canonical form. This is the same principle that lets HTML parsers handle tag-soup and makes Unix pipelines tolerate whitespace variations โ liberal acceptance enables evolution without breaking the old readers.
When to use each form
You are writing...
Use...
Why
inherits: frontmatter in a daughter skill
bare name (biome)
Short, resolver-driven, moves with the file
A prose see also: or related: in running text
unambiguous fragment (moollm/skills/biome/)
Helps readers open the right file without guessing
A cross-repo link in a design doc
unambiguous fragment
Makes the intent clear across repo boundaries
A relative link in a README.md that sits next to the referenced skill
relative path (../biome/)
Works in any Markdown renderer without a resolver
A citation that must survive the file being moved
unambiguous fragment
Paths starting with moollm/skills/ or <repo>/skills/ survive refactors
Disambiguation rules of thumb
If two skills in the same scope share a name, MOOLLM errors out rather than guessing. The author is expected to disambiguate using an unambiguous fragment.
A skill that deliberately shadows a more-distant one should say so in its own CARD.yml (shadows: moollm/skills/biome or similar) so readers understand the intent.
When writing consumer-repo skills that use MOOLLM skills by bare name, adding a comment to that effect (# Resolved to moollm/skills/biome/) is good practice but not required. The resolver should find it either way.
Empathic Templates โ FOOBAR.yml.template Files
A plural container directoryfoobars/ often ships with a companion file FOOBAR.yml.template. This is an empathic template โ a template that anticipates what the author of a new instance needs and carries that guidance inline.
Shape
foobars/ โ plural container
โโโ README.md โ explains what foobars are and how to add one
โโโ FOOBAR.yml.template โ empathic template (UPPERCASE because it declares the FOOBAR type)
โโโ foobar-alice/ โ instance: created by copying the template
โ โโโ FOOBAR.yml โ filled-in marker file
โ โโโ README.md
โโโ foobar-bob/
โโโ FOOBAR.yml
What makes a template "empathic"
An empathic template does three things an ordinary template doesn't:
Explains itself inline. Comments alongside every section say what the section is for, what typical values look like, what trade-offs to consider, and when to leave it blank.
Ships with sensible defaults. Fields carry default values that work for the most common case. The author can omit defaults they don't need to override, not just copy-and-edit.
Separates meta-commentary from payload. Blocks of explanation are clearly marked (# [INSTANTIATE: strip this section] or a # --- meta --- banner) so the author knows what to delete when finishing the instance.
Fill in required fields (values marked # REQUIRED or similar).
Override defaults only where you differ โ you can delete default-valued keys entirely; the template documents them so the instance file can stay minimal.
Strip the meta-commentary blocks. The instance is production documentation, not the template.
Keep the YAML comments that add authorial voice (per yaml-jazz). Comments explaining "why THIS instance is set up this way" are valuable and should stay. Comments of the form "โ replace with your value" are meta-commentary and should go.
Worked example
Template filefoobars/FOOBAR.yml.template:
# FOOBAR.yml.template โ empathic template for foobar instances# Copy this to foobars/<your-foobar-id>/FOOBAR.yml, fill in REQUIRED# fields, override defaults if you need to, delete this banner.# See foobars/README.md for the full instantiation recipe.foobar:id:my-foobar# REQUIRED โ kebab-case; must be unique within foobars/name:"My Foobar"# REQUIRED โ human-readable title# OPTIONAL โ defaults listed here; OMIT if you're fine with the default.flavor:standard# default: standard (options: standard, spicy, plain)region:auto# default: auto (detects from ambient context)enabled:true# default: true# REQUIRED โ this one has no reasonable default; every foobar differs.owner:null# e.g. "team-web"; pick an owner
Filled instancefoobars/foobar-alice/FOOBAR.yml โ written minimally, keeping only fields that differ from defaults or are required:
# FOOBAR.yml โ alice's foobar# Spicy flavor for the summer festival use-case; otherwise default.foobar:id:alicename:"Alice's Festival Foobar"flavor:spicyowner:team-festivals
Notice what the instance left out: region, enabled, and the template banner. Those use defaults and get inherited implicitly at read time.
Defaults resolution
A reader (human or LLM) merges the template's defaults with the instance's overrides:
This is Self-style prototype delegation applied to templates: the template is the prototype, the instance inherits from it. It composes naturally with DOP โ the prototype chain for a foobar instance is instance โ FOOBAR.yml.template โ biome/skill/whatever contains foobars/.
Why templates are declared UPPERCASE
FOOBAR.yml.template is UPPERCASE-prefixed because it declares the type of the things that go in foobars/. Same grammar as BIOME/ (the directory-scaffold template in skills/biome/templates/) โ the UPPERCASE names say "this is a type-level artifact, not a content file."
Two flavors of template in common use:
Template shape
Example
When to use
Single-file template
FOOBAR.yml.template or FOOBAR.md.template
Instance is a single file, or a directory with one marker file plus trivial extras
Directory scaffold
templates/FOOBAR/ containing FOOBAR.yml.tmpl, README.md.tmpl, etc.
Instance is a multi-file directory with several interfaces (e.g., a full skill with SKILL + CARD + GLANCE + README)
Both are valid. Use single-file when the instance is essentially one YAML document; use directory-scaffold when the instance exports multiple interfaces.
Templates as proto-skills
An empathic template is often the larval form of a skill. If the template starts getting used across unrelated contexts โ different parent skills, different repos, different domains โ lift it to a standalone skill:
The consumers that used the template in place switch to inherits: [foobar] and use the lifted skill's template directory. Same PLAY-LEARN-LIFT arc that governs skill promotion โ applied here to templates. A template that stays nested is one that hasn't yet earned the move.
Enumeration skip rules
Smart enumeration of a plural directory foobars/ should skip metadata and template files โ they're not instances. Per the grammar:
Filename pattern
Treat as
README.md, INDEX.md, INDEX.yml, ROOM.yml at container root
metadata (skip in enumeration)
*.yml.template, *.md.template
empathic template (skip)
*.yml.tmpl, *.md.tmpl
scaffolding template (skip)
*.yml.prototype, *.md.prototype
prototype (skip)
Any other entry with the expected UPPERCASE marker at its root
instance
Any other entry without the expected marker
auxiliary (skip or warn)
A smart queryInterface(dir, FOOBAR) returns null for any of the skipped entries โ they declare how to make foobars, not a foobar.
The Branch-as-Object Connection
Git branches are directories-with-history. The grammar extends:
Branch ALERT_42 in leela-ai/leela-alerts:
โโโ ALERT.yml โ "this branch exports the ALERT interface"
โโโ README.md โ "I have a human landing page"
โโโ attachments/
โโโ screenshot.png
The branch's root directory implements the ALERT interface, declared by ALERT.yml. The branch NAME (ALERT_42) encodes both the interface name (ALERT) and the instance id (42). Full protocol at central/skills/github/protocols/branch-as-object.md.
This is the same grammar operating in git-branch space instead of filesystem space. The rules don't change; only the storage backend does.
moollm/skills/schema/schemas/mechanisms/json-schema/
โโโ MECHANISM.yml โ "I export the MECHANISM interface โ I am a registered schema mechanism"
โโโ README.md โ "I have a human landing page"
Two interfaces, two marker files. That's the entire class.
moollm/skills/adventure/examples/adventure-4/characters/bartender/
โโโ CHARACTER.yml โ "I export the CHARACTER interface โ I am a character"
โโโ README.md
โโโ memory/ โ plural: contains memories
Practical Reading Protocol
When you encounter an unfamiliar MOOLLM or biome directory:
UPPERCASE marker files (they declare what this directory IS)
Collect the exports.ls <dir>/[A-Z]* gives you the interface list.
For each export, read the marker file at the appropriate resolution:
GLANCE.yml first if present (relevance check)
Then CARD.yml if you need the sniffable interface
Then SKILL.md / MECHANISM.yml / etc. for the full protocol
For each plural subdirectory, descend โ each of its children is a typed instance (probably with its own marker files).
For each singular subdirectory without a marker file, it's auxiliary โ not an instance, just organized content.
This is fast. It works at any depth. It scales from a 4-file skill to a 100+ directory tree.
Practical Authoring Protocol
When creating a new directory:
Decide: container or instance?
Container of type T โ name the directory <T-plural>/
Instance of type T โ name the directory singularly (id-form), add T.<ext> marker file at its root
Decide: which interfaces does this instance export?
For each interface you want to export, create a UPPERCASE marker file:
SKILL.md for a skill
CARD.yml for a card
GLANCE.yml for a glance
README.md for human landing
Your own FOO.md / FOO.yml for a new interface you're introducing
Do not register. There's no registry to update. The marker file IS the registration. (Unless you're in a skill like schema or biome that maintains a cross-cutting registry โ then yes, update that registry to point at your new instance, same as any member.)
Inherit if appropriate. In CARD.yml and/or GLANCE.yml and/or SKILL.md frontmatter, declare inherits: [parent-skill] to participate in DOP delegation.
Same grammar, git-branch storage backend. The branch root directory implements an interface declared by an UPPERCASE marker file.
Edge Cases & FAQs
Q: What if a directory has no UPPERCASE marker files?
Then it exports no interfaces under this grammar. It's "just a directory" โ maybe organizational (a subfolder of recipes, an index), maybe auxiliary (a tmp/ scratch dir, a snapshots/ archive). Parse accordingly.
Q: What about README.md? It's UPPERCASE.
README.md declares that this directory exports the README interface โ i.e., "I have a human-readable landing page." This is the lightest possible interface; nearly every MOOLLM directory exports it.
Q: What if the marker is in a subdirectory instead of the root?
Then the subdirectory is the exporter, not the parent. Marker files apply to their enclosing directory only. skills/gcs/gotchas/README.md declares that gcs/gotchas/ has a README, not that gcs/ has a README.
Q: What if I see .foo/ instead of foo/?
Dotfiles / dotdirs are hidden by convention (tools often skip them by default). Treat them as biome-specific configuration, not as interface exports. .github/ workflows is a good example.
Q: Is Dockerfile or Makefile an interface marker?
By our grammar, yes โ they're ALLCAPS / CamelCase marker files declaring that the directory exports a build interface. This existed long before MOOLLM; we're codifying a pre-existing convention. Same applies to LICENSE, CHANGELOG.md, CONTRIBUTING.md.
Q: What's the difference between skills/ and biomes/?
At the filesystem-object-grammar level, nothing โ both are plural containers declaring element type. The difference is semantic (skills are any MOOLLM skill; biomes are platform-descriptor skills that additionally inherit from the biome mother skill). The grammar is agnostic to domain; the inherits: chain carries the type refinement.
Q: Can one file export multiple interfaces?
Convention: no. One file โ one declared interface. If you want to export two, create two files. (Internally, a marker file may reference or include another, but each file's root existence is one declaration.)
Q: What about file-as-object, not just directory-as-object?
For content files (recipes, gotchas, runbooks, protocols), the plural parent dir declares the type; the file itself is the instance. Interface marker is implicit (the filename + parent-dir-type). This is the lighter form of the grammar โ directories get rich multi-interface treatment, content files get one-type-per-plural-container treatment.
Why This Matters
Without an explicit grammar:
Newcomers read MOOLLM directory trees as undifferentiated file piles.
LLMs navigate by keyword matching instead of type-aware traversal.
Authors reinvent conventions per skill; inconsistencies accumulate.
Cross-references break when directories restructure.
With the grammar:
A 5-second ls reveals what every directory IS and CONTAINS.
Interfaces are discoverable without registries, IDE plugins, or documentation scavenger hunts.
Authors add new interfaces by dropping a marker file; consumers discover by globbing.
Refactors are local โ move a directory, and its interface exports move with it.
The grammar is the file-system analog of COM's big insight ("multiple interfaces per class is the right abstraction") โ plus a human-readable naming scheme instead of UUIDs, plus the ls tool as a free type-checker.
Historical Lineage โ Where This Came From
This grammar did not appear from nowhere. The ideas it combines have been circling since the 1970s; MOOLLM is stapling them onto filesystems and LLM-legible contexts specifically. Oldest first:
Smalltalk (Xerox PARC, 1970s). Alan Kay et al. Classes and instances, message passing as the primitive mechanism, the image-as-object idea (your entire state is an object you can dump and reload). The spirit of "every thing is a kind of thing" lives here. MOOLLM's "directories are classes" reads as a Smalltalk image with the filesystem as the serialization.
Self (Ungar & Smith, 1987). The prototype revolution โ you don't need classes, just objects that delegate to parent objects. The delegation chain IS the type. MOOLLM's inherits: and PROTOTYPES.yml come directly from Self. Schemapedia's self mechanism profile has the full retrospective; the prototype skill is the MOOLLM-internal transmission of Self's ideas.
NeWS โ Network extensible Window System (James Gosling & David S. H. Rosenthal, Sun, 1986). PostScript-based window system. Lightweight prototype delegation proved in production system software, years before Self became widely known. NeWS dictionaries were Self-style objects before Self had a user base. The fact that you could build a full window system this way was the proof that prototype-based OO wasn't just a research curiosity.
TNT โ The NeWS Toolkit (Owen Densmore & David S. H. Rosenthal, Sun, late 1980s). A class hierarchy on top of NeWS's prototype foundation โ widgets, UI components, message dispatch โ organized by convention-naming that made class identity and inheritance legible as a tree. The patent Densmore and Rosenthal filed around this work is the direct conceptual ancestor of "treat the file system as an OO class hierarchy." MOOLLM's UPPERCASE-marker-file grammar is the filesystem-scoped reading of ideas TNT proved out at the widget level.
COM (Microsoft, mid-1990s). Multiple interfaces per implementation class โ the single idea MOOLLM borrows most directly in the "UPPERCASE marker file = exported interface" rule. QueryInterface(IID_IFoo, &ptr) maps to test -e <dir>/FOO.md. The UUIDs were COM's great mistake (opaque, registry-bound, Windows-centric); MOOLLM swaps them for UPPERCASE filenames.
XPCOM (Mozilla). Cross-platform COM; the "how to COM right" / "how COM went wrong" retrospective. Schemapedia's com-xpcom mechanism has the fuller history.
Ted Nelson โ intertwingularity (Project Xanadu, 1960s+). Explicit cross-reference is the primary mode. Filesystems work when cross-references are first-class navigation targets, not annotations on the side. MOOLLM's relentless cross-linking inherits this ethos.
MOOLLM DOP (../skill/delegation-object-protocol.md). Formalized Self-style delegation for LLM-read-able filesystems. PROTOTYPES.yml is an interface export under the grammar this skill names; DOP is the resolution algorithm that uses the grammar.
MOOLLM branch-as-object (leela-ai/leela-alerts Issue_ / ALERT_ branches). Extends the grammar from filesystems to git branches โ the branch's root directory declares its implemented interface via an UPPERCASE marker file. See central/skills/github/protocols/branch-as-object.md.
What MOOLLM adds to this long conversation:
LLM-legibility (the grammar is ls-readable and grep-friendly; no parser needed)
Semantic Image Pyramid (GLANCE โ CARD โ SKILL โ README as a multi-resolution reading order)
YAML comments as first-class semantic data (yaml-jazz)
Hierarchical nesting (<plural>/<singular>/<plural>/<singular>/...) natural and unbounded
Branch-as-object extension (git history as part of the object's timeline)
See Also
skill/SKILL.md โ MOOLLM meta-skill; uses this grammar to define what a skill IS
TNT: Densmore, O. & Rosenthal, D.S.H. "A user-interface toolkit in object-oriented PostScript." Computing Systems 3(1), Winter 1990. The paper that laid out the class hierarchy on PostScript NeWS dictionaries; widely cited as the patent-adjacent ancestor of "treat the filesystem as a class hierarchy."
COM:"Inside COM" โ Dale Rogerson, Microsoft Press, 1997. Or Don Box's "Essential COM", 1998 โ both are where "multiple interfaces per class" got popular.
