MCPInstall

metabase-static-embedding-to-guest-embedding-upgrade

Migrates a project from Metabase static embedding to guest embeds (web components via embed.js). Use when the user wants to migrate/convert/switch/upgrade from static embedding to guest embeds, from signed embed iframes to web components, or replace /embed/ iframes with metabase-dashboard/metabase-question components.

Installs
359
Last 24h
—
Trending rank
—
Security
Not audited

Install

Works with every agent the skills CLI supports.

npx skills add https://github.com/metabase/agent-skills --skill metabase-static-embedding-to-guest-embedding-upgrade
  1. 01

    Install

    Run the command in your project. The CLI asks which agents to install for.

  2. 02

    Describe the task

    Ask your agent for something this skill covers. It loads metabase-static-embedding-to-guest-embedding-upgrade automatically when relevant.

  3. 03

    Review the result

    The agent follows the SKILL.md steps, scripts and references bundled with the skill.

SKILL.md

The instructions your agent reads when the skill is triggered.

Execution contract

Follow the workflow steps in order — do not skip any step. Create the checklist first, then execute each step and explicitly mark it done with evidence. Each step's output feeds into the next, so skipping steps produces wrong migrations.

If you cannot complete a step due to missing info or tool failure, you must:

  1. record the step as ❌ blocked,
  2. explain exactly what is missing / what failed,
  3. stop (do not proceed to later steps).

Required output structure

Your response should contain these sections in this order:

  1. Step 0 Results: Metabase Version Detection
  2. Step 0.1: Migration Plan Checklist
  3. Step 1 Results: Project Scan + Docs Fetch
  4. Step 2 Results: Static Embed Analysis & Web Component Mapping
  5. Step 3: Migration Plan
  6. Step 4: Applied Code Changes
  7. Step 5: Validation
  8. Step 6: Final Summary

Each step section should end with a status line:

  • Status: ✅ complete or Status: ❌ blocked

Steps are sequential — do not start a step until the previous one is ✅ complete.

Evidence requirements

  • Step 0: Metabase version detected (source: Docker tag, env var, or user answer).
  • Step 1: every matched file path, every static embed location, JWT signing code, layout/head file, Metabase config variables, fetched docs listing.
  • Step 2: per embed — parsed iframe URL, content type, token variable, hash params, mapped web component with attributes.
  • Step 3: the complete file-by-file change plan with exact old/new code.
  • Step 4: per file — what was changed and exact diffs applied.
  • Step 5: each validation check's pass/fail result with evidence.

Architectural conformance

Follow the app's existing architecture, template engine, layout/partial system, code style, and route patterns. Do not switch paradigms (e.g., templates to inline HTML or vice versa). If the app has middleware for shared template variables, prefer that over duplicating across route handlers.

The web component must be rendered using the same delivery mechanism as the static iframe it replaces. If the iframe was rendered by a server-side template (EJS, Jinja, ERB, Blade, etc.), the web component should be rendered by the same template. If the iframe was returned as inline HTML from a route handler (e.g., res.send('<iframe ...')), the web component should be returned the same way. If the iframe was in a static HTML file, the web component goes in that same file. Do not move rendering from one layer to another — the migration should be a drop-in replacement at the same point in the rendering pipeline.

Token delivery must use the same mechanism as the original static embedding. If the JWT was rendered server-side into the HTML (e.g., res.send(\<iframe src=".../${token}">`)), the migrated web component should receive its token the same way — rendered server-side into the tokenattribute (e.g.,<metabase-dashboard token="${token}">). If the JWT was fetched client-side via fetch(), keep using fetch()` for the token. Do not change the delivery mechanism — just change what is delivered (raw token instead of full iframe URL).

Credential safety

This migration touches code that handles METABASE_SECRET_KEY and signed JWTs. Rules:

  • Never output literal secret values — if you encounter a hardcoded secret key or token string in the code, reference it by variable name only (e.g., "the secret in config.js:12"), never echo the value itself
  • In code diffs and summaries, use variable references (METABASE_SECRET_KEY, token) — not the resolved values
  • If you find hardcoded secrets, flag them to the user and recommend moving them to environment variables as part of the migration

Performance

  • Maximize parallelism within each step. Use parallel Grep/Glob/Read calls in single messages wherever possible.
  • Do not use sub-agents for project scanning — results need to stay in the main context for cross-referencing in later steps.
  • Do not parse repo branches, commits, PRs, or issues.

Scope

This skill converts static (signed) iframe embedding to guest embeds (web-component-based via embed.js). Both approaches use the same authentication model — signed JWTs with METABASE_SECRET_KEY — so the backend signing logic is preserved. The migration changes how the signed content is delivered: from iframes with JWT-in-URL to web components with a token attribute.

The consumer's app may be written in any backend language (Node.js, Python, Ruby, PHP, Java, Go, .NET, etc.) with any template engine. Keep instructions language-agnostic unless a specific language is detected in Step 1.

What this skill handles

  • Replacing signed <iframe> elements (/embed/dashboard/{JWT}, /embed/question/{JWT}) with web components (<metabase-dashboard token="...">, <metabase-question token="...">)
  • Adding the embed.js script tag (exactly once at app layout level)
  • Adding window.metabaseConfig with isGuest: true (exactly once at app layout level)
  • Mapping iframe hash parameters (#titled=true, #bordered=true) to web component attributes
  • Preserving the existing JWT signing logic — the backend still signs tokens with METABASE_SECRET_KEY using the same {resource, params} payload
  • Converting how the signed token reaches the frontend (from iframe URL path to template variable passed as token attribute)
  • Mapping locked params in the JWT to initial-parameters attribute where applicable
  • Removing iframeResizer.js references if present

What this skill does not handle

  • Migrating to SSO-based modular embedding (with user accounts) — this skill targets guest embedding only

How guest embeds differs from static iframe embedding

The auth model is the same — both use METABASE_SECRET_KEY to sign JWTs with {resource, params, exp}. What changes is how the embed is rendered:

AspectStatic embedding (iframe)Guest embeds (web component)
Element<iframe src="/embed/dashboard/{JWT}#params"><metabase-dashboard token="{JWT}">
Token deliveryBaked into iframe URL pathPassed as token attribute
ConfigNone (iframe is self-contained)window.metabaseConfig = { isGuest: true, instanceUrl: "..." }
ScriptOptional iframeResizer.jsRequired embed.js
AppearanceHash params (#titled=true)Component attributes (with-title="true")
Locked paramsIn JWT params fieldSame JWT params field (unchanged)
Secret keyMETABASE_SECRET_KEYSame METABASE_SECRET_KEY

Guest embeds support additional attributes (e.g., downloads, drill-through, hidden parameters) not available in static embedding. Consult the fetched docs for the full list of available attributes for the target version.

Allowed documentation sources

Fetch the version-specific llms-embedding-full.txt using this URL:

https://www.metabase.com/docs/v0.{VERSION}/llms-embedding-full.txt

The version in the URL uses the format v0.58 (normalize: strip leading v or 0., drop patch — e.g., 0.58.1 → 58 → URL uses v0.58). This single file contains all embedding documentation for that version, optimized for LLM consumption.

Other constraints:

  • No GitHub PRs/issues or npm pages
  • Do not follow changelog links to GitHub or guess URLs

AskUserQuestion triggers

Use AskUserQuestion and halt until answered if:

  • The Metabase instance URL cannot be determined from project code or environment variables
  • The backend language cannot be determined
  • The Metabase instance version cannot be determined from the project code
  • No layout/head file can be identified (unclear where to inject embed.js)
  • Multiple layout files exist and it is unclear which one(s) to use

Pre-workflow steps

Migration Plan Checklist

Create a checklist to track progress. In Claude Code, use TaskCreate/TaskUpdate tools:

  • Step 0: Detect Metabase version
  • Step 1: Scan project + fetch target version docs
  • Step 2: Analyze static embeds and map to web components (using docs)
  • Step 3: Plan migration changes
  • Step 4: Apply code changes
  • Step 5: Validate changes
  • Step 6: Final summary

Workflow

Step 0: Detect Metabase version

Before anything else, determine the Metabase version. Grep the project for Docker image tags (metabase/metabase:v, metabase/metabase-enterprise:v), METABASE_VERSION, or version references. If undetected, AskUserQuestion (options: v53 or older, v54–v58, v59+). Abort if < v53 (modular embedding not available). Record the version.

Step 1: Scan the project + fetch docs

Perform the project scan and doc fetch concurrently — they are independent. Use parallel tool calls within a single message wherever there are no dependencies.

1a: Fetch target version docs

Fetch llms-embedding-full.txt for the target version (see "Allowed documentation sources" for URL format). These docs are the authoritative source for web component attributes, window.metabaseConfig options, and guest embedding configuration for the target version. Use them in Step 2 for mapping instead of relying on hardcoded tables alone.

Launch this concurrently with the project scan steps below.

1b: Identify backend language and framework

  • Check for dependency/build files (package.json, requirements.txt, Gemfile, pom.xml, go.mod, composer.json, etc.).
  • Identify the template engine and record the language and framework.

1c: Check for existing modular embedding setup

Grep for these patterns (in parallel) to detect if the app already has modular embedding configured:

  • /app/embed.js — existing embed.js script tag
  • window.metabaseConfig — existing config assignment

Record whether each is already present and where. If both already exist (e.g., the app uses modular embedding alongside static embedding), Steps 3a and 3b will skip adding them.

1d: Find ALL static embedding code

Use Grep to search for all of these patterns (in parallel):

  • /embed/dashboard/ in all files — static embed dashboard URLs
  • /embed/question/ in all files — static embed question URLs
  • <iframe in all template/HTML/JSX/view files — the embed elements
  • METABASE_SECRET_KEY or METABASE_EMBEDDING_SECRET_KEY — the signing secret
  • resource: near dashboard or question — JWT payload structure
  • iframeResizer — optional auto-resize script

For each file with a match, read the entire file.

1e: Find JWT signing code

Use Grep to search for all of these patterns (in parallel):

  • jwt.sign or jwt.encode or JWT or jsonwebtoken or PyJWT or jose
  • METABASE_SECRET_KEY or MB_EMBEDDING_SECRET_KEY
  • resource: combined with params: (the static embed JWT payload shape)

For each matching file, read the entire file.

1f: Find the layout/head file(s)

Find the single file (or common code path) where the HTML <head> section is defined — this is where embed.js and window.metabaseConfig will be injected (unless already present per Step 1c).

Search for:

  • <head> or <!DOCTYPE or <html in template/view files
  • Layout/wrapper patterns: include('head'), <%- include, {% extends, {% block, layout, base.html, _layout, application.html
  • If the app builds HTML via inline strings in server code (e.g., res.send(...)), identify where the <head> content is generated

1g: Find Metabase configuration

Grep for METABASE_ and MB_ prefixed variables. Record every Metabase-related variable name and where it is read.

Output: Structured Project Inventory

Compile all findings into:

Backend: {language}, {framework}, {template engine}
Metabase config:
  - Site URL variable: {name} (read at {file}:{line})
  - Secret key variable: {name} (read at {file}:{line})
  - Other variables: ...
Layout/head file: {path}:{line range}
Static embeds found: {count}
  - {file}:{line} — {brief description} (dashboard/question, ID: {id})
  - ...
JWT signing: {file}:{line} — {library used}
JWT payload: resource type={dashboard|question}, params={list or "none"}
iframeResizer: {present|not present}
Existing modular embedding: {embed.js: yes/no, metabaseConfig: yes/no}

Step 2: Analyze static embeds and map to web components (ONLY after Step 1 ✅)

Use the documentation fetched in Step 1a as the authoritative reference for web component attributes, window.metabaseConfig options, and guest embedding behavior. The hardcoded tables below are fallbacks — if the docs describe additional attributes or different behavior for the target version, prefer the docs.

For each static embed found in Step 1:

2a: Parse the signed iframe URL

Extract from the iframe src attribute:

  • Metabase base URL: may come from env var, constant, or be hardcoded
  • Resource type: dashboard or question (from the /embed/{type}/ path)
  • Resource ID: the numeric ID from the JWT resource field (e.g., resource: { dashboard: 10 })
  • Locked parameters: any params in the JWT payload (e.g., params: { category: ["Gadget"] })
  • Hash parameters: appearance customization after # (e.g., #titled=true&bordered=false)
  • iframeResizer usage: whether iFrameResize() is called on this iframe

2b: Map content type to web component

Static embed URL patternModular Web ComponentRequired Attribute
/embed/dashboard/{JWT}<metabase-dashboard>token="{JWT}"
/embed/question/{JWT}<metabase-question>token="{JWT}"

The token attribute receives the same signed JWT that was previously baked into the iframe URL. The backend signing code stays the same — only the delivery mechanism changes.

If the token was built dynamically in a template (e.g., src="<%= metabaseUrl %>/embed/dashboard/<%= token %>"), extract the token variable and pass it as the token attribute (e.g., token="<%= token %>").

2c: Map hash parameters

Parameters that map to web component attributes:

Static embedding hash paramsGuest embeds equivalent
titled=true/falsewith-title="true/false" on the component
bordered=true/falseNo direct equivalent — drop (web components have no border chrome)
refresh=NNo direct equivalent — drop (handled by Metabase instance config)
theme=nightUse window.metabaseConfig.theme instead (if supported by version)

2d: Map locked and editable parameters

Locked parameters (in JWT params field) — no change needed. They remain in the JWT and continue to work the same way. The signed token already contains them.

Editable parameters — if the static embed allowed users to interact with filters, these can now be set as defaults via the initial-parameters attribute:

<metabase-dashboard
  token="{JWT}"
  initial-parameters='{"category":["Doohickey","Gizmo"]}'
></metabase-dashboard>

initial-parameters sets default filter values that the user can change. This is a new capability not available in static iframe embedding.

2e: Output Migration Mapping Table

For each static embed, output:

embed #{n}: {file}:{line}
  Old: {iframe HTML or signing + iframe code}
  Content type: {dashboard|question}
  Token variable: {template expression for the signed JWT}
  Locked params: {in JWT — no change needed}
  Hash params: {list or "none"}
  Dropped params: {list}
  Mapped attributes: {list}
  New: {exact replacement web component HTML}

Step 3: Plan migration changes (ONLY after Step 2 ✅)

Create a complete file-by-file change plan covering all areas below. Every change should be specified with the target file, the old code, and the new code.

3a: metabaseConfig — exactly once per app

Skip this step if Step 1c found an existing window.metabaseConfig assignment. If it exists but is missing isGuest: true, add that field to the existing config instead of creating a new one.

  • Target: the layout/head file identified in Step 1f
  • Location: inside <head>, before the embed.js script tag (the config must be set before embed.js loads)
  • Code to add:
    <script>
      window.metabaseConfig = {
        isGuest: true,
        instanceUrl: "{METABASE_SITE_URL}",
      };
    </script>
    
  • isGuest: true is required — it tells embed.js to use guest (signed token) mode instead of SSO mode.
  • instanceUrl should be rendered dynamically using the project's template expression syntax.
  • Locale: If a locale parameter was found in any static embed hash, add locale: "{code}" to the config object.
  • Consult the fetched docs (Step 1a) for any additional window.metabaseConfig options supported by the target version (e.g., theme, font).
  • window.metabaseConfig should be set exactly once.

3b: embed.js script injection — exactly once per app

Skip this step if Step 1c found an existing embed.js script tag.

  • Target: same layout/head file as 3a
  • Location: inside <head>, after the window.metabaseConfig script (embed.js reads the config on load)
  • Code to add:
    <script defer src="{METABASE_SITE_URL}/app/embed.js"></script>
    
  • {METABASE_SITE_URL} should be rendered dynamically using the project's existing template expression syntax.
  • Verify this will appear exactly once in the rendered HTML regardless of which page the user visits.

3c: Refactor backend token delivery

The backend already has JWT signing code that produces the token. Currently it builds a full iframe URL (/embed/dashboard/{token}#params). The signing logic stays — but how the token reaches the frontend changes:

  • Before: Backend builds full iframe URL string, passes to template, template renders <iframe src="{url}">
  • After: Backend passes just the signed token to the template, template renders <metabase-dashboard token="{token}">

For each signing location found in Step 1d:

  1. Keep the JWT signing call (jwt.sign(payload, METABASE_SECRET_KEY)) unchanged
  2. Remove the URL construction code that prepended {baseUrl}/embed/dashboard/ and appended hash params
  3. Pass the raw token string to the template context instead of the full URL

If the signing happens inline in the template handler (not in a shared function), the change is local to that handler.

3d: iframe replacement plan

For EACH iframe from Step 2e's Migration Mapping Table:

  • Specify: file path, exact old code to replace, exact new code
  • The new web component uses token="{token_variable}" where {token_variable} is the template expression for the signed JWT
  • Map hash parameters to component attributes per Step 2c
  • Preserve styling: Transfer the iframe's sizing directly to the web component element — no wrapper <div> needed:
    • If the iframe had width/height HTML attributes or inline style, apply them directly to the web component (e.g., <metabase-dashboard token="..." style="width:800px;height:600px">)
    • If the iframe was styled via CSS classes, apply those classes directly to the web component
    • If the iframe was inside a container that already controls sizing, no extra styling needed — the web component will fill that container
    • If the iframe used iframeResizer for auto-height, drop it — web components handle their own sizing
  • Remove any iframeResizer calls associated with this iframe

3e: Dead code removal

After replacing iframes, identify and remove:

  • URL construction code that built /embed/dashboard/{token}#params or /embed/question/{token}#params strings
  • iframeResizer.js script tag and any iFrameResize() calls
  • Hash parameter string construction (e.g., const mods = "titled=true&bordered=false")
  • Any helper functions that were only used for building static embed iframe URLs

Do not remove:

  • JWT signing code (jwt.sign(payload, METABASE_SECRET_KEY)) — still used
  • METABASE_SECRET_KEY env var — still used
  • JWT library imports — still used
  • Any code used by other parts of the application

3f: Metabase admin configuration notes (manual steps for the user)

List these as part of the plan — they will be included in the final summary:

  1. Enable modular embedding: Admin > Embedding > toggle "Enable modular embedding"
  2. Enable guest embedding: Admin > Embedding > ensure "Guest embedding" (or "Static embedding" in older UI) is enabled. The existing static embedding secret key is reused.
  3. Configure CORS origins: Admin > Embedding > Security > add the host app's domain (e.g., http://localhost:9090). This is new — static iframe embedding did not require CORS configuration.

Step 4: Apply code changes (ONLY after Step 3 ✅)

Apply all changes from Step 3 in this order:

  1. First: Add window.metabaseConfig assignment and embed.js script tag to the layout/head file (Step 3a + 3b, config before embed.js)
  2. Second: Refactor backend token delivery — keep signing, remove URL construction (Step 3c)
  3. Third: Replace each iframe with its web component (Step 3d), one file at a time
  4. Fourth: Remove dead code — iframeResizer, URL builders (Step 3e)

Constraints:

  • Use the Edit tool with precise old_string / new_string for every change
  • Do not add new package dependencies — modular embedding requires only the embed.js script served by the Metabase instance
  • Do not change or remove METABASE_SECRET_KEY — it is still used for signing
  • If a file requires multiple edits, apply them top-to-bottom to avoid offset issues

Step 5: Validate changes (ONLY after Step 4 ✅)

Perform all of these checks. Each check should have an explicit pass/fail result.

5a: No remaining static embed iframes

Use Grep to search for /embed/dashboard/ and /embed/question/ across all project files (excluding node_modules, .git, lockfiles).

Pass criteria: zero static embed URL constructions found (the pattern may still appear in comments — verify these are not live code).

5b: embed.js appears exactly once

Use Grep to search for /app/embed.js across all project files (excluding node_modules, .git).

Pass criteria: exactly ONE occurrence in the layout/head file.

5c: window.metabaseConfig is set exactly once

Use Grep to search for window.metabaseConfig across all project files (excluding node_modules, .git).

Pass criteria: exactly ONE occurrence with isGuest: true.

5d: JWT signing code is preserved

Read the JWT signing file(s). Verify:

  • jwt.sign (or equivalent) call still exists
  • METABASE_SECRET_KEY is still read from environment
  • JWT payload still contains resource and params fields

Pass criteria: signing logic intact.

5e: No remaining iframeResizer references

Use Grep to search for iframeResizer and iFrameResize across all project files.

Pass criteria: zero references remain (or only in unrelated code).

5f: Spot-check modified files

Read each modified file and verify:

  • Web components have token attribute with correct template expression
  • Template syntax is valid (no unclosed tags, correct expressions)
  • Dead code identified in Step 3e has been removed

Pass criteria: all checks pass.

If ANY check fails:

  • Fix the issue immediately
  • Re-run the specific check
  • If unable to fix after 3 attempts, mark Step 5 ❌ blocked and report which check failed and why

Step 6: Output summary

Organize the final output into these sections:

  1. Changes applied: list every file modified and a one-line description of each change
  2. Web component mapping: table showing each old signed iframe → new web component:
    | File | Old | New |
    |---|---|---|
    | views/analytics.ejs | <iframe src="/embed/dashboard/{token}#titled=true"> | <metabase-dashboard token="{token}" with-title="true"> |
    
  3. What stayed the same: JWT signing logic, METABASE_SECRET_KEY, locked parameters in JWT params field
  4. Dropped parameters: list of static embed hash parameters that were dropped, with brief explanation
  5. New capabilities available: features now accessible that weren't in static iframe embedding:
    • initial-parameters attribute for editable filter defaults
    • with-downloads attribute for enabling downloads (Pro/Enterprise)
    • Better mobile responsiveness (web components adapt to container)
  6. Manual steps required (Metabase admin configuration from Step 3f):
    • Enable modular embedding
    • Ensure guest embedding is enabled (reuses existing secret key)
    • Configure CORS origins (new requirement)
  7. Behavioral differences the user should be aware of:
    • Web components expand to fill their container — if the iframe had fixed dimensions, verify the container provides appropriate sizing
    • The bordered appearance option is no longer available — web components render without a frame
    • Auto-refresh (refresh=N) is no longer controlled per embed — configure it in Metabase instance settings instead

Retry policy

Doc fetching:

  • If fetching llms-embedding-full.txt returns 404, verify the Metabase version number and retry. If still failing, mark Step 1 ❌ blocked.

Validation:

  • If any validation check in Step 5 fails after 3 fix attempts, mark Step 5 ❌ blocked and report which check failed and why.
  • If AskUserQuestion is not answered, remain blocked on that step — do not guess or proceed with assumptions.

FAQ

What does the metabase-static-embedding-to-guest-embedding-upgrade skill do?

Migrates a project from Metabase static embedding to guest embeds (web components via embed.js). Use when the user wants to migrate/convert/switch/upgrade from static embedding to guest embeds, from signed embed iframes to web components, or replace /embed/ iframes with metabase-dashboard/metabase-question components.

How do I install metabase-static-embedding-to-guest-embedding-upgrade?

Run npx skills add https://github.com/metabase/agent-skills --skill metabase-static-embedding-to-guest-embedding-upgrade in your project. The skills CLI detects the agents you use and copies the skill into the right folder. For Claude Code you can also place the skill folder at ~/.claude/skills/metabase-static-embedding-to-guest-embedding-upgrade/ (personal) or .claude/skills/metabase-static-embedding-to-guest-embedding-upgrade/ (project).

Which AI agents can use metabase-static-embedding-to-guest-embedding-upgrade?

Any agent that supports the open Agent Skills format (a folder with a SKILL.md file), including Claude Code, Codex, Cursor, GitHub Copilot, Gemini CLI, Windsurf, Cline, OpenCode and more.

Is metabase-static-embedding-to-guest-embedding-upgrade safe to install?

No partner security audit is available yet. Skills can run commands and scripts, so read the SKILL.md and any bundled files before installing.

Who maintains metabase-static-embedding-to-guest-embedding-upgrade?

It is published by metabase in metabase/agent-skills. It has 359 recorded installs.

More from metabase/agent-skills

View all →

rde

metabase/agent-skills

Expert data engineering with Metabase at the center of any company's stack: explore and profile raw warehouse tables, build clean tables as Metabase transforms pinned by transform tests, build the semantic layer (models, metrics, measures, segments, metadata), design dashboards, answer questions with checked numbers, reconcile against a reference, change delivered definitions safely. Use for data work, not one mb command (metabase-cli): "make sense of my data", "model this raw schema", "set up analytics for X", "build a data model", "define MRR / active customers officially", "build a semantic layer", "go from raw tables to a dashboard", "does this number match finance", "our numbers look wrong", "two cards disagree", "test this transform", "why does this transform get case X wrong", "make Metabot answer questions about X", "explain this table", "change this definition", "build a data app", "donor retention", "event attendance", "quarterly board report", "be my data engineer / anal

Data & Databases949 installs

rde-data-apps

metabase/agent-skills

Build or change a Metabase data app on an rde machine: settles what the metabase-data-apps skill needs from the local setup (the remote-sync repository, the Metabase URL, the API key rde stores, how the app reaches Metabase), makes sure everything the app reads is Library content already synced into that repository, confirms every sync step with the user, and hands the build to it. Use only when an app is asked for explicitly: "build me an app for X", "create a data app", "make an internal tool / portal in Metabase", "add a page / a form / a filter to my data app". A dashboard, a question, or a metric is not an app; those stay with the rde skill.

Backend & APIs877 installs

metabase-cli

metabase/agent-skills

Drive a Metabase instance from the terminal via the mb CLI: auth profiles, databases and schemas, schema sync and field-values rescan, tables, fields, cards (questions, models, metrics) run as JSON/CSV/XLSX, dashboards and dashcards, collections and their tree, snippets, segments, measures, transforms and transform-jobs, settings, search, git-sync. Use for any Metabase operation from the terminal: "log into metabase", "what profiles do I have", "list cards", "run card 42 as CSV", "create a transform", "list dashboards", "move a dashcard", "list collections", "what's in collection 4", "show the collection tree", "list snippets", "create a segment", "archive a measure", "search metabase for X", "import the latest changes", "add a directory to git sync", "set a setting", "what schemas are in this database", "trigger a sync", "rescan field values", or any mb . Whole data jobs go to rde.

Data & Databases784 installs

metabase-database-metadata

metabase/agent-skills

Understands the Metabase Database Metadata Format — a YAML-based on-disk representation of databases, tables, and fields synced from a Metabase instance. Use when the user needs to read, edit, or understand metadata files produced by @metabase/database-metadata, or when reasoning about a project's schema (columns, types, FK relationships) through the .metadata/databases folder.

Data & Databases564 installs

metabase-representation-format

metabase/agent-skills

Understands the Metabase Representation Format — a YAML-based serialization format for Metabase content (collections, cards, dashboards, documents, segments, measures, snippets, transforms, actions, and data apps). Use when the user needs to create, edit, understand, or validate Metabase representation YAML files, or when working with Metabase serialization/deserialization (serdes). Covers entity schemas, MBQL and native queries, visualization settings, parameters, and folder structure.

AI & Agents535 installs

metabase-semantic-checker

metabase/agent-skills

Runs the Metabase semantic checker against a tree of Representation Format YAML files to verify that all references resolve — cross-entity references (collectionid, dashboardid, parentid, parameter source cards, snippet references, transform tags, etc.) and references to columns inside MBQL and native queries. Slow (≥1 min per run). Only use when the user explicitly asks to verify entity references or column references in MBQL/SQL queries; in most cases this runs as a CI step, not locally. Requires database metadata on disk (by default .metadata/tablemetadata.json).

Data & Databases434 installs

Popular ai & agents skills

View category →

google-agents-cli-publish

google/agents-cli

This skill should be used when the user wants to "publish an agent", "publish my ADK agent", "register an agent with Gemini", "publish to Gemini" (formerly Gemini Enterprise), or needs guidance on the agents-cli publish gemini command. Also use when the user wants to "manage agents in Agent Registry", "list/update/delete registered agents", or "register an MCP server". Covers ADK vs A2A registration modes, programmatic and interactive usage, flag reference, auto-detection from deployment metadata, Agent Registry fleet management, and troubleshooting. Part of the agents-cli skills suite. Do NOT use for deployment (use google-agents-cli-deploy).

AI & Agents513.9K installs

google-agents-cli-eval

google/agents-cli

This skill should be used when the user wants to "run an evaluation", "evaluate my agent", "evaluate my ADK agent", "write an eval dataset", "analyze eval failures", "compare eval results", "optimize agent", or needs guidance on the Gemini platform eval methodology and the Quality Flywheel. Covers eval metrics, dataset schema, LLM-as-judge scoring, and common failure causes. Applies to any agents-cli project, whatever framework the agent is written in. Do NOT use for agent API code patterns (ADK: use google-agents-cli-adk-code), deployment (use google-agents-cli-deploy), or project scaffolding (use google-agents-cli-scaffold).

AI & Agents513.5K installs

skill-creator

anthropics/skills

Create new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, edit, or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy.

OfficialAI & Agents416.4K installs

codex-pet

prime-skills/runcomfy-agent-skills

Codex Pet generator on RunComfy. Build a Codex-compatible Codex Pet spritesheet.webp + pet.json from a single reference image, drop it into ${CODEXHOME:-$HOME/.codex}/pets/ / and Codex picks it up as a custom Codex Pet next to the 8 built-ins. This skill produces the exact Codex Pet atlas Codex expects (1536x1872 PNG/WebP, 8 cols x 9 rows, 192x208 cells, 9 animation states — idle, running-right, running-left, waving, jumping, failed, waiting, running, review). Calls OpenAI GPT Image 2 edit ONCE via the local RunComfy CLI as runcomfy run openai/gpt-image-2/edit to produce a canonical Codex Pet pose, then assembles all 9 animation rows programmatically with ImageMagick micro-transforms — no Codex Pro, no $imagegen, no OPENAIAPIKEY required, only RUNCOMFYTOKEN. Triggers on "codex pet", "create codex pet", "make codex pet", "hatch codex pet", "/hatch image", "desktop pet codex", "codex pets", "spritesheet.webp", or any explicit ask to build a custom pet for OpenAI Codex.

AI & Agents402.1K installs

full-output-enforcement

leonxlnx/taste-skill

Overrides default LLM truncation behavior. Enforces complete code generation, bans placeholder patterns, and handles token-limit splits cleanly. Apply to any task requiring exhaustive, unabridged output.

AI & Agents401K installs

executing-plans

obra/superpowers

Use when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available

AI & Agents245.7K installs