v0.1.0 Apache-2.0 · MCP-native · open
Skill Wiki.
Skill Wiki turns each SKILL.md blob into a
wiki of typed atoms — addressable by ID, linked
by a declared edge graph, retrieved on demand. Agents see a ~3 KB
index of what exists; atom bodies stream in only when the brief
actually calls for them.
Each @scope is a Prime — a self-contained corpus on
the open Prime protocol that
powers Skill Wiki.
Why Skill Wiki
Stop bulk-loading SKILL.md files into your agent.
Three architectural inversions, each with a measurable cost saved. Read the background for the long version.
Existence ≠ content
Agents always see a ~3 KB _index.xml of what exists. Atom
bodies load only when retrieval picks them — addressed by ID, not guessed
by relevance.
Typed edges, not flat related
14 verbs (requires, validates-with,
contradicts, specializes, …) carry semantics the
parser and L3 checker reason over. The graph isn't decoration.
Three projection levels
Each atom compiles to summary (~30 tok),
core (~150 tok), full (~380 tok). Pick the
smallest projection that fits the budget.
Spec surface · v1.0 frozen
The shape of the wiki
Four numbers fixed at v1.0 of the Prime protocol — the spec the wiki is built on. They don't change when you add more Primes.
— and what's currently shipping —
Marketplace · live
Reference deployment
Numbers from the 6 reference Primes in this site's marketplace — compiled with the v0.1.0 reference implementation.
See the full listing in Marketplace →
Flagship Prime
From the marketplace
Skill Wiki is domain-blind — actual knowledge lives in the Primes.
Each Prime is a self-contained @scope of typed atoms —
what Anthropic's SKILL.md format would look like if every
line had a type. Today's marketplace already spans design, backend
architecture, and application security.
@frontend-design is the flagship reference Prime:
898 typed atoms about visual
design — the deepest worked example of the protocol on a real domain.
@frontend-design
898 typed atoms across 9 design sub-domains — color, typography, motion, spacing, accessibility, copy voice, hierarchy, layout, and component patterns. Reference corpus + reference application: 5-tool MCP wrapper, 31 personas, 30 task taxonomies, 20-task benchmark fixtures.
- Atoms
- 898
- Kinds
- 28 of 28
- Tokens
- 422,867 compiled
- Namespaces
- 5
- License
- Apache-2.0
@backend-architecture
@backend-architecture v0.1.0Domain plugin for backend / distributed-system architecture corpora. Covers 12-Factor App principles, microservices patterns, and the layer / deployment / scale axes that classify the trade-offs.
@security-appsec
@security v0.1.0Domain plugin for application-security corpora derived from the OWASP Cheat Sheet Series. Covers input validation, authentication, session management, cryptography, injection prevention, XSS, CSRF, secrets management, and core security principles.
@web-style
@web-style v0.1.0Domain plugin for JavaScript style corpora. Covers naming conventions, variable declarations, function shape, object and array idioms, module layout, and core readability principles drawn from Airbnb and Google style guides.
@android-compose
@android-compose v0.1.0Domain plugin for Jetpack Compose component API guidance and Material 3 design tokens. Covers composable function shape, state hoisting, modifiers, slot APIs, recomposition discipline, lazy lists, Material 3 color/typography/elevation/motion tokens, accessibility, and adaptive layout patterns.
@ios-swift
@ios-swift v0.1.0Domain plugin for Swift / iOS / Apple-platform corpora derived from the official Swift API Design Guidelines and the Google Swift Style Guide. Covers naming, protocol- and value-oriented design, optional handling, error propagation, memory management in closures, and structured concurrency.
Get started
Install in 30 seconds.
Skill Wiki ships as an MCP server. Point Claude Code, Cursor, Continue, or any MCP-capable client at it.
1 · Boot the server
bun add -g @skill-wiki/cli
bunx @skill-wiki/mcp-server-core 2 · Wire into Claude Code
{
"mcpServers": {
"skill-wiki": {
"command": "bunx",
"args": ["@skill-wiki/mcp-server-core"],
"env": { "PRIME_DIR": "/abs/path/to/compiled" }
}
}
}