Carmen LabsCarmen Labs
MView V1

Install the MView skill

Instructions for an LLM to install a skill for creating and validating MViews from the canonical sources.

Install the MView skill

This page is intended for an LLM with access to the user's filesystem. Its purpose is to install a reusable skill for creating, reviewing, and validating MViews without duplicating a copy of the specification that may become outdated.

The skill must use Create an MView with an AI agent as its entry point. That page is designed for models: it explains the creation workflow and links to the specific documentation that must be consulted at each step.

Instructions for the installing model

When the user asks you to install the MView skill, perform these actions:

  1. Identify the skills directory supported by the agent you are configuring. Inspect the project's instructions and configuration first. Prefer a project-local installation so it remains portable with the repository.
  2. Create a directory named mview inside that skills directory.
  3. Create mview/SKILL.md with the template from this page.
  4. If a skill with that name already exists, read it before editing. Update only the MView instructions and preserve local adaptations that do not conflict with the canonical sources.
  5. Do not download or maintain a local copy of the JSON Schema as the source of truth. The skill must consult the canonical URL to avoid becoming outdated.
  6. Verify that SKILL.md exists at the expected location and report the installed path.

Common locations include .agents/skills/mview/SKILL.md, .claude/skills/mview/SKILL.md, and .opencode/skills/mview/SKILL.md, but you must use the actual convention of the agent and project instead of assuming one of them.

SKILL.md template

Install the following content:

---
name: mview
description: Create, review, and validate MViews from source files or selected files in a source directory using the canonical Carmen Labs MView specification. Use when the user asks to create, modify, inspect, or validate an MView, mview.json, MView parser, or MView component.
---
 
# MView
 
Use the current Carmen Labs documentation as the source of truth. Do not rely on remembered field names, accepted values, or copied schema definitions.
 
## Canonical sources
 
- Model-oriented creation guide: https://carmenlabs.com/v1/create
- Specification overview: https://carmenlabs.com/v1/overview
- Manifest reference: https://carmenlabs.com/v1/schema-reference
- Canonical JSON Schema: https://schemas.carmenlabs.com/mview/v1/schema.json
- Capabilities and normalized input: https://carmenlabs.com/v1/capabilities
- Parser contract: https://carmenlabs.com/v1/parser-contract
- Component contract: https://carmenlabs.com/v1/component-contract
- Directory and discovery rules: https://carmenlabs.com/v1/directory-layout
- Security model: https://carmenlabs.com/v1/security-model
- Versioning: https://carmenlabs.com/v1/versioning
 
## Required workflow
 
1. Fetch and read `https://carmenlabs.com/v1/create` before creating an MView. Treat it as the navigation entry point for models, then follow the linked specification pages needed for the task.
2. Inspect the source file or the explicitly selected files in a source directory locally. Determine their structure, fields, encodings, missing values, and the user's intended view. Never invent source columns or data, and never include undeclared directory files implicitly.
3. Read the capabilities page and select `accepts` from the normalized input the parser actually needs, not only from the source file extension.
4. Read the manifest reference and fetch the canonical JSON Schema before writing `mview.json`. The schema is the authority for fields, required properties, enums, patterns, and `additionalProperties`.
5. Read the parser and component contracts before implementing them. Keep source-file I/O in the runtime, make the parser deterministic, and make the component render only the model it receives.
6. Read the directory-layout and security pages before choosing paths or adding runtime behavior. All manifest paths must be relative.
7. Preserve the user's original creation request in `mview.json` as `prompt`.
8. Validate the finished manifest against `https://schemas.carmenlabs.com/mview/v1/schema.json`. Also verify that parser output matches component input and that every declared file exists.
 
## Source-data discovery
 
The creation guide tells you which documentation to consult; it does not contain the user's source data. Find data only through files or tools the user has made available in the current workspace:
 
1. Locate the source file or directory named by the user. For a directory, identify the exact selected files. If the source or selection is ambiguous or unavailable, ask for its path or for a representative sample.
2. Inspect only enough data to infer the real structure and safely implement the parser.
3. Do not upload source data to Carmen Labs documentation URLs. Those URLs provide specification material only.
4. Set `source.path` relative to the MView directory. For a directory, set `source.kind` to `"directory"`, make every `source.files[].path` relative to it, and declare each file's `accepts`. Use content hashes only when they were calculated according to the canonical manifest reference.
 
## Non-negotiable rules
 
- Never modify a source file as part of MView execution.
- Never let a directory MView read files outside its explicit `source.files` selection.
- Never use absolute paths in the manifest.
- Never add manifest properties that are absent from the current schema.
- Never let the parser or component bypass the runtime to read arbitrary files.
- Never substitute `metadata` for rendering configuration; values that affect rendering belong in `settings`.
- Prefer the simplest renderer supported by the target runtime and justified by the requested view.
 
If a canonical URL cannot be accessed, state which source could not be verified. Do not guess the current contract or claim schema conformance without validation.

How the skill finds information

The skill does not need to store the entire specification. When activated, it:

  1. Opens /v1/create to obtain the model-oriented workflow.
  2. Follows links from that page according to the task: capabilities for interpreting input, the schema reference for the manifest, and contracts for the parser and component.
  3. Fetches the schema directly from schemas.carmenlabs.com/mview/v1/schema.json before validation.
  4. Inspects actual data only in the workspace or through tools authorized by the user.

This separation keeps the skill small and allows it to always consult the published version of the specification.

Installation check

Installation is complete when:

  • Exactly one mview/SKILL.md exists in the selected skills directory.
  • Its frontmatter contains name: mview and a description that activates the skill for MView tasks.
  • The creation guide and JSON Schema are listed as canonical sources.
  • The workflow requires inspecting real data, consulting the documentation, and validating the manifest.
  • The agent can discover the skill after starting a new session or reloading its skill catalog.

Instalar mview skill

Instrucciones para que un modelo LLM instale una skill que le permita crear y validar MViews usando las fuentes canónicas.

Instalar mview skill

Esta página está dirigida a un modelo LLM con acceso al sistema de archivos del usuario. Su objetivo es instalar una skill reutilizable para crear, revisar y validar MViews sin duplicar una copia de la especificación que pueda quedar obsoleta.

La skill debe usar Crear una MView con un agente de IA como punto de entrada. Esa página está preparada para modelos: explica el flujo de creación y enlaza la documentación específica que se debe consultar en cada paso.

Instrucciones para el modelo instalador

Cuando el usuario te pida instalar la skill de MView, realiza estas acciones:

  1. Identifica la carpeta de skills compatible con el agente que estás configurando. Revisa primero las instrucciones y la configuración del proyecto. Prefiere una instalación local al proyecto para que sea portable con el repositorio.
  2. Crea una carpeta llamada mview dentro de esa carpeta de skills.
  3. Crea mview/SKILL.md con el contenido de la plantilla de esta página.
  4. Si ya existe una skill con ese nombre, léela antes de editarla. Actualiza solo las instrucciones de MView y conserva cualquier adaptación local que no contradiga las fuentes canónicas.
  5. No descargues ni mantengas una copia local del JSON Schema como fuente de verdad. La skill debe consultar la URL canónica para evitar desactualizaciones.
  6. Verifica que SKILL.md exista en la ubicación esperada y comunica la ruta instalada.

Ubicaciones habituales incluyen .agents/skills/mview/SKILL.md, .claude/skills/mview/SKILL.md y .opencode/skills/mview/SKILL.md, pero debes usar la convención real del agente y del proyecto en vez de asumir una de ellas.

Plantilla de SKILL.md

Instala el siguiente contenido:

---
name: mview
description: Create, review, and validate MViews from source files or selected files in a source directory using the canonical Carmen Labs MView specification. Use when the user asks to create, modify, inspect, or validate an MView, mview.json, MView parser, or MView component.
---
 
# MView
 
Use the current Carmen Labs documentation as the source of truth. Do not rely on remembered field names, accepted values, or copied schema definitions.
 
## Canonical sources
 
- Model-oriented creation guide: https://carmenlabs.com/v1/create
- Specification overview: https://carmenlabs.com/v1/overview
- Manifest reference: https://carmenlabs.com/v1/schema-reference
- Canonical JSON Schema: https://schemas.carmenlabs.com/mview/v1/schema.json
- Capabilities and normalized input: https://carmenlabs.com/v1/capabilities
- Parser contract: https://carmenlabs.com/v1/parser-contract
- Component contract: https://carmenlabs.com/v1/component-contract
- Directory and discovery rules: https://carmenlabs.com/v1/directory-layout
- Security model: https://carmenlabs.com/v1/security-model
- Versioning: https://carmenlabs.com/v1/versioning
 
## Required workflow
 
1. Fetch and read `https://carmenlabs.com/v1/create` before creating an MView. Treat it as the navigation entry point for models, then follow the linked specification pages needed for the task.
2. Inspect the source file or the explicitly selected files in a source directory locally. Determine their structure, fields, encodings, missing values, and the user's intended view. Never invent source columns or data, and never include undeclared directory files implicitly.
3. Read the capabilities page and select `accepts` from the normalized input the parser actually needs, not only from the source file extension.
4. Read the manifest reference and fetch the canonical JSON Schema before writing `mview.json`. The schema is the authority for fields, required properties, enums, patterns, and `additionalProperties`.
5. Read the parser and component contracts before implementing them. Keep source-file I/O in the runtime, make the parser deterministic, and make the component render only the model it receives.
6. Read the directory-layout and security pages before choosing paths or adding runtime behavior. All manifest paths must be relative.
7. Preserve the user's original creation request in `mview.json` as `prompt`.
8. Validate the finished manifest against `https://schemas.carmenlabs.com/mview/v1/schema.json`. Also verify that parser output matches component input and that every declared file exists.
 
## Source-data discovery
 
The creation guide tells you which documentation to consult; it does not contain the user's source data. Find data only through files or tools the user has made available in the current workspace:
 
1. Locate the source file or directory named by the user. For a directory, identify the exact selected files. If the source or selection is ambiguous or unavailable, ask for its path or for a representative sample.
2. Inspect only enough data to infer the real structure and safely implement the parser.
3. Do not upload source data to Carmen Labs documentation URLs. Those URLs provide specification material only.
4. Set `source.path` relative to the MView directory. For a directory, set `source.kind` to `"directory"`, make every `source.files[].path` relative to it, and declare each file's `accepts`. Use content hashes only when they were calculated according to the canonical manifest reference.
 
## Non-negotiable rules
 
- Never modify a source file as part of MView execution.
- Never let a directory MView read files outside its explicit `source.files` selection.
- Never use absolute paths in the manifest.
- Never add manifest properties that are absent from the current schema.
- Never let the parser or component bypass the runtime to read arbitrary files.
- Never substitute `metadata` for rendering configuration; values that affect rendering belong in `settings`.
- Prefer the simplest renderer supported by the target runtime and justified by the requested view.
 
If a canonical URL cannot be accessed, state which source could not be verified. Do not guess the current contract or claim schema conformance without validation.

Cómo encuentra la información la skill

La skill no necesita almacenar toda la especificación. Al activarse:

  1. Abre /v1/create para obtener el flujo orientado a modelos.
  2. Sigue los enlaces de esa página según la tarea: capacidades para interpretar la entrada, referencia del esquema para el manifiesto y contratos para el parser y el componente.
  3. Obtiene el esquema directamente desde schemas.carmenlabs.com/mview/v1/schema.json antes de validar.
  4. Inspecciona los datos reales únicamente en el workspace o mediante las herramientas autorizadas por el usuario.

Esta separación mantiene la skill pequeña y permite que siempre consulte la versión publicada de la especificación.

Comprobación de la instalación

La instalación está completa cuando:

  • Existe un único mview/SKILL.md en la carpeta de skills seleccionada.
  • El frontmatter contiene name: mview y una descripción que activa la skill para tareas de MView.
  • La guía de creación y el JSON Schema aparecen como fuentes canónicas.
  • El workflow exige inspeccionar datos reales, consultar la documentación y validar el manifiesto.
  • El agente puede descubrir la skill al iniciar una nueva sesión o al recargar su catálogo de skills.