Skill Catalog

A skill is the method layer: a document telling the expert how this kind of work is done, usually bundled with executable scripts and templates. Tools decide what an expert can do; skills decide how well it does it. This page enumerates the built-in skills and explains the three ownership tracks.

Three ownership tracks (the key concept)

The skills available in one conversation are the union of three sources, never overriding each other:

Track	Qualified prefix	Storage	Visible in
Built-in	`builtin/*`	In the image, `services/agent-runtime/src/plugins/builtin/*/skills/`	Every conversation
User global	`user/*`	`user_skills` table, keyed by user ID	That user’s conversations with any expert
Expert-scoped	`expert/*`	`expert_skills` table, keyed by expert identifier	That expert’s conversations only

Same names do not collide: builtin/docx, user/docx, and expert/docx can coexist. What the expert sees in skill discover is the qualified name, and activation must use it in full — a bare name that matches multiple tracks is refused rather than guessed.

Write entry points are strictly partitioned by ownership:

Entry point	Writes to
Installing from the skill store / My Skills	`user_skills` (global library)
Installing in expert configuration	`expert_skills`
Expert calling `skill_install` in a conversation	`expert_skills`, and it requires an expert context
Marketplace expert publish	`expert_skills`, inside the transaction

Conversations without an expert context — free chat, IM channels — refuse skill_install and point you at the UI to install into the global library. This is deliberate: a skill with no owning object has nowhere to be filed. Built-in skills never enter the database. Changing them means changing the image and shipping a release.

How a skill gets used

discover

The expert calls skill(action="discover") and receives the list — names plus descriptions only, no bodies loaded.

activate

Having judged one relevant, it calls skill(action="activate", name="builtin/pptx") to read the full body into context.

execute

For scripts and templates it calls skill(action="execute", resource_type="script", resource_name="..."). Scripts run black-box: only stdout/stderr returns, never the source.

The two-stage design (catalog first, body second) is mandatory: in the table below builtin/pptx is 34,079 characters and builtin/kami is 31,269. Injecting everything would consume most of a context window outright.

Built-in skills

Grouped by plugin. Character counts are skill-document lengths, useful for judging the context cost of activating one.

Document processing

Qualified name	Characters	Purpose
`builtin/docx`	21,208	Creating, reading, editing Word documents
`builtin/xlsx`	11,455	Any task where a spreadsheet is the primary input or output
`builtin/pdf`	8,424	PDF reading, extraction, forms, merge/split
`builtin/kami`	31,269	Professional typesetting: resumes, one-pagers, white papers, portfolios
`builtin/pptx`	34,079	Presentation hub, routing to native pptx or HTML deck
`builtin/html-deck`	22,221	Horizontal-swipe web PPT as a single HTML file, two style templates

pptx and kami are the two longest documents on the platform because they carry complete production specs (templates, script invocations, export locks).

Visualization (22 skills)

Qualified name	Characters	Purpose
`builtin/visualize`	1,911	Master skill: inline HTML/CSS/JS visualizations in chat
`builtin/visualization-strategy`	2,523	Choosing a chart type from data and intent
`builtin/data-visualization`	1,847	Routes to the right data-viz sub-skill
`builtin/d3-data-visualization`	2,744	D3.js bar, line, scatter, custom
`builtin/statistical-visualization`	2,840	Histograms, box plots, violin plots, regression
`builtin/grammar-of-graphics`	2,755	Vega-Lite / Observable Plot, declarative
`builtin/canvas2d-data-visualization`	2,308	10K+ point datasets where SVG chokes
`builtin/dashboards-realtime`	3,122	Live dashboards, KPI cards, sparklines
`builtin/gantt-chart`	3,429	Timelines and Gantt charts
`builtin/node-link-diagram`	3,342	Network graphs, dependency trees, force-directed, chord
`builtin/uml-architecture`	3,099	UML, architecture, sequence diagrams
`builtin/geospatial-visualization`	2,521	D3-geo, choropleth, TopoJSON
`builtin/scrollytelling`	3,005	Scroll-driven narrative visualizations
`builtin/molecular-visualization`	3,850	Proteins, nucleic acids, ligands, PDB/mmCIF
`builtin/interactive-3d-atlas`	6,902	Clickable-hotspot 3D atlases (anatomical, mechanical, architectural)
`builtin/threejs-showcase`	15,577	Five battle-tested Three.js patterns
`builtin/threejs-data-visualization`	3,266	3D bars, globes, 3D networks, terrain
`builtin/3d-data-visualization`	7,485	Mapping data into 3D space
`builtin/react-nextjs-visualization`	2,728	Embedding in React/Next.js components
`builtin/a2ui-patterns`	5,910	A2UI layout patterns (comparisons, dashboards, wizards)
`builtin/accessibility-visualization`	2,134	ARIA, keyboard navigation, colorblind-safe palettes
`builtin/reports-pdf-slides`	3,749	Export to PDF, slides, print-ready formats

The profy-visualize manifest declares only 8 skills while the directory holds 22. That is not an error: the 8 declared ones are injected into the catalog when the plugin activates, and the other 14 are discovered on demand via discover then activate. It prevents one plugin tick from dumping 22 entries into context.

Game development (10 skills)

Qualified name	Characters	Purpose
`builtin/game-studio`	888	Routes to a sub-skill by project type
`builtin/web-game-foundations`	1,123	Architecture patterns shared by all web games
`builtin/three-webgl-game`	2,357	Vanilla Three.js + WebGL 3D games
`builtin/react-three-fiber-game`	1,654	R3F + Drei 3D games
`builtin/phaser-2d-game`	2,380	2D games with Phaser 3
`builtin/physics-game`	8,544	Rapier WASM physics games (pinball, marble run, tower stacking)
`builtin/game-ui-frontend`	1,200	HUD, menus, dialogs over the game canvas
`builtin/sprite-pipeline`	1,023	Creating and animating 2D sprite assets
`builtin/web-3d-asset-pipeline`	1,251	Optimizing 3D assets for browser games
`builtin/game-playtest`	2,060	Automated playtesting via screenshots and state assertion

Video and creative

Qualified name	Characters	Purpose
`builtin/movie` (`skills/SKILL.md`)	4,611	Video production pipeline overview
`builtin/video-production`	2,020	Seven-section production methodology template
`builtin/product-ad`	1,727	Product ads (15–60s, high energy, CTA-focused)
`builtin/animated-short`	1,843	Animated shorts (1–3 min, narrative-driven)
`builtin/world-generation`	2,188	Navigable 3D worlds with HY-World 2.0
`builtin/canvas`	2,801	Canvas workflow: create nodes, wire the DAG, auto-layout, run

Desktop and professional software bridges

Qualified name	Characters	Purpose
`builtin/blender-workflow`	1,652	Blender modeling best practices
`builtin/godot-workflow`	2,015	Godot development with deterministic playtesting
`builtin/photoshop-workflow`	1,284	Non-destructive Photoshop editing discipline
`builtin/chrome-extension`	3,514	Driving the user’s authenticated Chrome

The three software-bridge skills are all short (1,284–2,015 characters) because they carry workflow discipline only; the concrete operations depend on MCP tools — and that part is currently unwired. See the known-defect section on each plugin page.

Record & Replay (9 skills, flat files)

profy-record-and-replay stores its skills as flat .md files rather than directories:

File	Characters	Purpose
`record-replay.md`	32,194	Main record-and-replay document
`cua-driver.md`	13,027	Desktop control driver
`cua-recording.md`	2,770	Recording conventions
`app-clock.md`	2,845	Clock app adaptation
`app-notion.md`	2,323	Notion adaptation
`app-music.md`	1,385	Music app adaptation
`app-numbers.md`	1,089	Numbers adaptation
`app-spotify.md`	980	Spotify adaptation
`app-iphone-mirroring.md`	415	iPhone Mirroring adaptation

The app-* series is a per-application operation manual — “click play” has a different accessibility tree in every app, so each one has to be recorded individually.

Platform mechanics

Qualified name	Characters	Purpose
`builtin/skill`	20,383	How to write an effective skill (used when a user wants to author one)
`builtin/distill`	17,634	Distillation conversation methodology
`builtin/knowledge`	14,480	Search and write across connected knowledge platforms
`builtin/sites`	13,104	Deep reference for the Sites pipeline
`builtin/evolution`	5,374	Evolution-mode scoring anchors and workflow
`builtin/customs-extraction`	8,177	83-field customs declaration extraction

builtin/evolution and builtin/distill are only visible in their respective internal modes; a normal conversation never sees them.

Writing a skill

A skill is a directory containing at minimum a SKILL.md:

my-skill/
├── SKILL.md          # Required: frontmatter + body
├── scripts/          # Executable scripts (black-box execution)
├── references/       # Reference material (read on demand)
├── templates/        # Template files
└── assets/           # Static assets

The SKILL.md frontmatter:

---
name: my-skill
description: One line on when this skill applies. This sentence is the only thing that reaches the discover list
---

description decides whether the skill gets remembered at all. During discover the expert sees nothing else — a brilliant body behind a vague description is never activated. Write the triggering condition, not a self-introduction.Compare against the table above: builtin/xlsx says “any task where a spreadsheet is the primary input or output” — that is a trigger. Not “Excel processing capability,” which is an introduction.

Context cost

A skill body enters context in full on activate. Rough scale:

Length	Impact
< 3,000 characters	Negligible
3,000–10,000	Normal; two or three can coexist
10,000–20,000	Noticeable; activate one at a time
> 20,000	A single skill occupies a meaningful share; needs active restraint

This is why self-evolution breaks out context_footprint as its own scoring dimension computed by code rather than self-scored by the model: longer and more detailed usually scores higher when the model grades itself, but the real cost is paid on every single call.

Skill system

Concepts and usage

Tool Catalog

Full signatures for skill and skill_manage

Plugin Catalog

Which plugin ships which skill

Self-evolution

How skills improve automatically

Verified 2026-08-11. Sources: services/agent-runtime/src/plugins/builtin/*/skills/ (47 directory-style skills, 7 flat SKILL.md files, 9 flat record-and-replay files). Character counts are measured document lengths.

Get Started

Core Concepts

Guides

Platform & Account

Deep Dive

Skill Catalog

Skill Catalog

Three ownership tracks (the key concept)

How a skill gets used

Built-in skills

Document processing

Visualization (22 skills)

Game development (10 skills)

Video and creative

Desktop and professional software bridges

Record & Replay (9 skills, flat files)

Platform mechanics

Writing a skill

Context cost

Skill system

Tool Catalog

Plugin Catalog

Self-evolution

​Skill Catalog

​Three ownership tracks (the key concept)

​How a skill gets used

​Built-in skills

​Document processing

​Visualization (22 skills)

​Game development (10 skills)

​Video and creative

​Desktop and professional software bridges

​Record & Replay (9 skills, flat files)

​Platform mechanics

​Writing a skill

​Context cost

​Related pages

Skill system

Tool Catalog

Plugin Catalog

Self-evolution

Skill Catalog

Three ownership tracks (the key concept)

How a skill gets used

Built-in skills

Document processing

Visualization (22 skills)

Game development (10 skills)

Video and creative

Desktop and professional software bridges

Record & Replay (9 skills, flat files)

Platform mechanics

Writing a skill

Context cost

Related pages