Package as a Cursor plugin (skill + Context7 MCP)

.cursor-plugin/marketplace.json

@@ -0,0 +1,15 @@
+{
+  "name": "iterable",
+  "owner": { "name": "Iterable" },
+  "metadata": {
+    "description": "Iterable's official skills for AI coding assistants — mobile SDK integration guidance.",
+    "version": "0.1.0"
+  },
+  "plugins": [
+    {
+      "name": "iterable-sdk",
+      "source": "./",
+      "description": "Iterable mobile SDK skill"
+    }
+  ]
+}

.cursor-plugin/plugin.json

@@ -0,0 +1,10 @@
+{
+  "name": "iterable-sdk",
+  "version": "0.1.0",
+  "description": "Reliable Iterable mobile SDK integration guidance for AI coding assistants — version-pinned snippets + silent-failure pitfalls. Android today; iOS/RN/Web later.",
+  "author": { "name": "Iterable" },
+  "repository": "https://github.com/Iterable/iterable-sdk-skill",
+  "license": "MIT",
+  "keywords": ["iterable", "android", "sdk", "mobile", "push", "in-app"],
+  "skills": ["./iterable-android"]
+}

.github/workflows/validate-plugins.yml

@@ -0,0 +1,57 @@
+name: Validate plugin manifests
+
+# Lightweight gate for Cursor/Claude plugin packaging — no Java/Kotlin needed.
+
+on:
+  push:
+    branches: [main]
+    paths:
+      - .cursor-plugin/**
+      - .claude-plugin/**
+      - mcp.json
+      - .mcp.json
+      - pipeline/src/validate-plugins.ts
+      - pipeline/package.json
+      - .github/workflows/validate-plugins.yml
+  pull_request:
+    branches: [main]
+    paths:
+      - .cursor-plugin/**
+      - .claude-plugin/**
+      - mcp.json
+      - .mcp.json
+      - pipeline/src/validate-plugins.ts
+      - pipeline/package.json
+      - .github/workflows/validate-plugins.yml
+  workflow_dispatch:
+
+permissions:
+  contents: read
+
+jobs:
+  validate-plugins:
+    runs-on: ubuntu-latest
+    timeout-minutes: 5
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v4
+
+      - name: Setup pnpm
+        uses: pnpm/action-setup@v4
+        with:
+          version: 9
+
+      - name: Setup Node
+        uses: actions/setup-node@v4
+        with:
+          node-version: 20
+          cache: pnpm
+          cache-dependency-path: pipeline/pnpm-lock.yaml
+
+      - name: Install pipeline deps
+        working-directory: pipeline
+        run: pnpm install --frozen-lockfile
+
+      - name: Validate plugin manifests
+        working-directory: pipeline
+        run: pnpm validate:plugins

.mcp.json

@@ -2,7 +2,10 @@
   "mcpServers": {
     "context7": {
       "type": "http",
-      "url": "https://mcp.context7.com/mcp"
+      "url": "https://mcp.context7.com/mcp",
+      "headers": {
+        "CONTEXT7_API_KEY": "${env:CONTEXT7_API_KEY}"
+      }
     }
   }
 }

README.md

@@ -21,14 +21,30 @@ callback, and more).
 
 This installs the skill and wires up its documentation source in one step.
 
-**Cursor (or a manual install)** — symlink the skill directory into your
-assistant's skills folder:
+**Cursor** — install it as a plugin, which wires up the skill **and** its
+documentation source (the Context7 MCP server) in one step. Cursor 3.9+ supports
+Agent Skills and plugins natively.
+
+Clone the repo and symlink it into Cursor's local-plugins folder, then restart
+Cursor (or reload the window):
 
 ```bash
-git clone --depth 1 git@github.com:Iterable/iterable-sdk-skill.git ~/iterable-skills
-ln -s ~/iterable-skills/iterable-android ~/.cursor/skills/iterable-android
+git clone --depth 1 https://github.com/Iterable/iterable-sdk-skill.git ~/iterable-skills
+ln -s ~/iterable-skills ~/.cursor/plugins/local/iterable-sdk
 ```
 
+A Cursor team admin can instead publish it to the whole org via **Dashboard →
+Plugins → Add Marketplace → Import from Repo**, pointing at this repo.
+
+> **Just the skill, no plugin?** Cursor 3.9+ auto-loads Agent Skills from
+> `~/.cursor/skills/` — symlink only the skill directory:
+> `ln -s ~/iterable-skills/iterable-android ~/.cursor/skills/iterable-android`.
+> This skips the doc source; to add it, paste this into your browser to install
+> the Context7 MCP server:
+> ```
+> cursor://anysphere.cursor-deeplink/mcp/install?name=context7&config=eyJ1cmwiOiJodHRwczovL21jcC5jb250ZXh0Ny5jb20vbWNwIn0=
+> ```
+
 Either way, the skill activates automatically whenever you mention Iterable: it
 loads its always-on rules and `PITFALLS.md`, then pulls the documentation for
 whatever feature you're working on (see [How it works](#how-it-works)).
@@ -39,15 +55,16 @@ The skill carries a copy of the Iterable documentation inside it
 (`iterable-android/snapshot/`), so it always has the docs on hand — even offline.
 This snapshot is the active source today.
 
-The Claude Code plugin also connects to [Context7](https://context7.com), a
-service that hosts docs for AI assistants to query on demand. That connection is
-bundled and ready, but stays dormant until Iterable's curated library is
-published there; once it is, the skill fetches the latest docs live, with the
-snapshot as its fallback. No reinstall needed when that happens.
+The plugin (both Claude Code and Cursor) also connects to
+[Context7](https://context7.com), a service that hosts docs for AI assistants to
+query on demand. That connection is bundled and ready, but stays dormant until
+Iterable's curated library is published there; once it is, the skill fetches the
+latest docs live, with the snapshot as its fallback. No reinstall needed when
+that happens.
 
 > Context7 works without an API key at a lower rate limit. To raise it, get a
-> free key at [context7.com](https://context7.com) and add it as a
-> `CONTEXT7_API_KEY` header on the `context7` MCP server.
+> free key at [context7.com](https://context7.com) and set the `CONTEXT7_API_KEY`
+> environment variable — the bundled MCP config forwards it automatically.
 
 ## What it covers
 
@@ -67,6 +84,10 @@ polished/           the docs in agent-ready form (what gets published to Context
 pipeline/           tooling that builds polished/ from sources/, CI-gated
 sources/            raw Iterable docs, fetched at pinned commits
 context7.json       Context7 manifest
+.claude-plugin/     Claude Code plugin + marketplace manifests
+.cursor-plugin/     Cursor plugin + marketplace manifests
+mcp.json            Context7 MCP server (Cursor plugin auto-discovery)
+.mcp.json           same config (Claude Code auto-discovery; kept in sync by CI)
 ```
 
 ## Staying current

iterable-android/SKILL.md

@@ -33,10 +33,12 @@ human readers but omit several silent-failure traps documented in
 Do this **first**, before Preflight and before any edits. The SDK covers many
 features — push, in-app, mobile inbox, embedded, deep links, event tracking,
 user profiles. Do **not** assume the developer wants all of them, and do **not**
-start implementing on a guessed scope. Ask which features they want via
-`AskUserQuestion`. Set `multiSelect: true` so they can pick several, and offer
-**at most 4 options** (the tool rejects more than 4 per question) — group the
-long tail under one bucket, e.g.:
+start implementing on a guessed scope. Ask which features they want. If your
+host exposes an interactive multi-select question tool (in Claude Code,
+`AskUserQuestion` with `multiSelect: true`; in Cursor, `AskQuestion` with
+`allow_multiple: true`), use it and offer **at most 4 options** (those tools
+reject more than 4 per question); otherwise ask in plain text. Either way, group
+the long tail under one bucket, e.g.:
 
 - Push notifications (FCM)
 - In-app messages
@@ -88,17 +90,19 @@ pauses for a missing input is **always better** than one that compiles by
 faking the input — the latter ships a broken or misleading state that looks
 done.
 
-> **Always ask via selectable options, not prose — every time.** For *any*
-> question to the developer — the Step 0 scope question, the inputs below, the
-> identity model, region, and "what would you like to do next?" — use the
-> `AskUserQuestion` tool so they get interactive choices, not a plain-text list
-> they must answer by typing. Offer realistic options with a short description
-> each (e.g. identity: "Stable per-install UUID via `setUserId`" / "Account
-> email via `setEmail`"; region: "US" / "EU"). **Offer at most 4 options per
-> question** — the tool rejects more than 4 and the call fails with an
-> "invalid parameters" error. If you have more than 4, group them or split into
-> a second question. Only fall back to plain text if the question genuinely has
-> no enumerable options.
+> **Prefer selectable options over prose — every time.** For *any* question to
+> the developer — the Step 0 scope question, the inputs below, the identity
+> model, region, and "what would you like to do next?" — if your host exposes an
+> interactive question tool (in Claude Code, `AskUserQuestion`; in Cursor,
+> `AskQuestion`), use it so they get interactive choices instead of a plain-text
+> list they must answer by typing. Offer realistic options with a short
+> description each (e.g. identity: "Stable per-install UUID via `setUserId`" /
+> "Account email via `setEmail`"; region: "US" / "EU"). With those tools,
+> **offer at most 4 options per question** — they reject more than 4 and the
+> call fails with an "invalid parameters" error; if you have more than 4, group
+> them or split into a second question. If no such tool is available, ask in
+> plain text with the options clearly enumerated. Only skip options entirely if
+> the question genuinely has none.
 
 | Input | Needed when | If missing |
 |---|---|---|

mcp.json

@@ -0,0 +1,11 @@
+{
+  "mcpServers": {
+    "context7": {
+      "type": "http",
+      "url": "https://mcp.context7.com/mcp",
+      "headers": {
+        "CONTEXT7_API_KEY": "${env:CONTEXT7_API_KEY}"
+      }
+    }
+  }
+}

pipeline/package.json

@@ -10,6 +10,7 @@
     "fetch:sources": "tsx src/fetch.ts",
     "polish:a": "tsx src/polish-layer-a.ts",
     "validate:polished": "tsx src/validate-polished.ts",
+    "validate:plugins": "tsx src/validate-plugins.ts",
     "check:snippets": "tsx src/check-snippets.ts",
     "lint:chunking": "tsx src/lint-chunking.ts",
     "recompute:manifest": "tsx src/recompute-manifest.ts",
@@ -20,7 +21,7 @@
     "eval:run": "tsx src/eval-run.ts run",
     "eval:report": "tsx src/eval-report.ts",
     "typecheck": "tsc --noEmit",
-    "check:all": "pnpm typecheck && pnpm validate:polished && pnpm snapshot:verify && pnpm check:snippets && pnpm lint:chunking"
+    "check:all": "pnpm typecheck && pnpm validate:polished && pnpm validate:plugins && pnpm snapshot:verify && pnpm check:snippets && pnpm lint:chunking"
   },
   "devDependencies": {
     "@types/node": "^22.7.0",