stackcanvas 0.2.0 → 0.2.2

package/README.md

@@ -1,7 +1,5 @@
 # stackcanvas
 
-[Telemetry: opt-in, anonymous, no infra data — see TELEMETRY.md](TELEMETRY.md)
-
 Live infrastructure canvas for [Claude Code](https://claude.com/claude-code).
 The agent writes and plans your Terraform — stackcanvas shows it as a living
 diagram. Drag new resources onto the canvas; the agent turns them into
@@ -16,7 +14,7 @@ localhost, reading your local state and plan.
 2. The graph re-renders live whenever `*.tfstate` or `.stackcanvas/plan.json`
    change — you watch the agent work.
 3. You drag resources from the palette (or right-click existing ones to
-   request changes / removal) and hit **Apply**. Connections between two
+   request changes / removal) and hit **Send to agent**. Connections between two
    not-yet-created (draft) resources aren't included in the intent yet —
    connect drafts to existing resources, or describe the relation in the
    draft's wishes field.
@@ -30,10 +28,6 @@ localhost, reading your local state and plan.
     claude plugin marketplace add pshenok/stackcanvas
     claude plugin install stackcanvas@stackcanvas
 
-or without the plugin system:
-
-    claude mcp add stackcanvas -- npx -y stackcanvas
-
 Then, inside a repo with Terraform:
 
     /stackcanvas
@@ -41,6 +35,13 @@ Then, inside a repo with Terraform:
 This is the verified path — the CI `check-plugin` job validates the plugin
 and marketplace manifests on every push.
 
+Or without the plugin system:
+
+    claude mcp add stackcanvas -- npx -y stackcanvas
+
+Then, inside a repo with Terraform, just ask: *open the stackcanvas canvas
+for this repo*.
+
 ## Other MCP clients
 
 stackcanvas is a standard [MCP](https://modelcontextprotocol.io) server, so
@@ -77,11 +78,14 @@ What's provider-specific is only the curation layer, shipped as **provider packs
 
 - a palette pack (`packages/ui/src/resource-palette.ts`) — curated drag-and-drop types
 - containment rules (`DEFAULT_CONTAINMENT_RULES` in `@stackcanvas/core`) — which
-  resources render as visual containers (AWS VPC/subnet today)
+  resources render as visual containers (AWS VPC/subnet, GCP network/subnetwork,
+  Azure subnet, Cloudflare zone today)
 - icon patterns (`packages/ui/src/icons.tsx`)
 
-AWS ships first. Adding a GCP/Azure/other pack is pure data and makes a great
-first PR.
+Four packs ship today: **AWS** (complete — the reference pack) and **GCP /
+Azure / Cloudflare starter packs** covering the common resource types per
+provider. Rounding out a starter pack, or adding a new provider entirely, is
+pure data and makes a great first PR.
 
 ## OpenTofu
 
@@ -98,7 +102,7 @@ to pin it — both take precedence over auto-detection.
 | `open_canvas` | Start the canvas for a Terraform root, open the browser |
 | `load_plan` | Register a plan (JSON or binary) for diff highlighting |
 | `get_graph_summary` | Text summary of the graph for the agent |
-| `await_canvas_intent` | Block until the user clicks Apply; returns their edits |
+| `await_canvas_intent` | Block until the user clicks Send to agent; returns their edits |
 
 ## Demo
 
@@ -106,13 +110,14 @@ to pin it — both take precedence over auto-detection.
 
 ## Telemetry
 
-stackcanvas can send a handful of anonymous usage counters (installs,
-canvases opened, edits sent, scans run) — **opt-in only**, nothing is sent
-until you click **Allow** on the one-time canvas banner, and `DO_NOT_TRACK=1`
-/ `STACKCANVAS_TELEMETRY=0` always turn it off regardless of that decision.
-No resource names, infrastructure data, or file paths ever leave your
-machine. Full payload, consent model, and how to verify it yourself:
-[TELEMETRY.md](TELEMETRY.md).
+stackcanvas can send a handful of anonymous (pseudonymous install id — see
+TELEMETRY.md) usage counters (installs, canvases opened, intents sent;
+scan/drift counters reserved — five ever, see TELEMETRY.md) — **opt-in
+only**, nothing is sent until you click **Allow** on the one-time canvas
+banner, and `DO_NOT_TRACK=1` / `STACKCANVAS_TELEMETRY=0` always turn it off
+regardless of that decision. No resource names, infrastructure data, or file
+paths ever leave your machine. Full payload, consent model, and how to
+verify it yourself: [TELEMETRY.md](TELEMETRY.md).
 
 ## Development
 

package/dist/cli.js

@@ -40,7 +40,15 @@ function deriveEdges(nodes) {
 }
 var DEFAULT_CONTAINMENT_RULES = [
   { containerType: "aws_vpc", memberAttr: "vpc_id", kind: "vpc" },
-  { containerType: "aws_subnet", memberAttr: "subnet_id", kind: "subnet" }
+  { containerType: "aws_subnet", memberAttr: "subnet_id", kind: "subnet" },
+  // GCP networks/subnetworks are referenced by `self_link`, not `id`.
+  { containerType: "google_compute_network", memberAttr: "network", kind: "vpc", containerIdAttr: "self_link" },
+  { containerType: "google_compute_subnetwork", memberAttr: "subnetwork", kind: "subnet", containerIdAttr: "self_link" },
+  { containerType: "azurerm_subnet", memberAttr: "subnet_id", kind: "subnet" },
+  // Cloudflare zones aren't a network container, but they group the
+  // resources scoped to them the same way a VPC groups its members — reuse
+  // the 'vpc' kind rather than introducing a one-off styling kind.
+  { containerType: "cloudflare_zone", memberAttr: "zone_id", kind: "vpc" }
 ];
 function deriveContainment(g, rules = DEFAULT_CONTAINMENT_RULES) {
   const groups = [...g.groups];
@@ -48,7 +56,7 @@ function deriveContainment(g, rules = DEFAULT_CONTAINMENT_RULES) {
   for (const rule of rules) {
     const groupByPhysicalId = /* @__PURE__ */ new Map();
     for (const n of nodes.filter((n2) => n2.type === rule.containerType)) {
-      const pid = n.attributes["id"];
+      const pid = n.attributes[rule.containerIdAttr ?? "id"];
       if (typeof pid !== "string") continue;
       const gid = `${rule.kind}:${n.id}`;
       groups.push({ id: gid, label: n.name, kind: rule.kind, parent: n.group });
@@ -926,7 +934,7 @@ function openBrowser(url) {
 }
 
 // src/version.ts
-var VERSION = "0.2.0";
+var VERSION = "0.2.2";
 
 // src/server.ts
 function looksLikeTerraformRoot(dir) {
@@ -966,7 +974,11 @@ function createMcpServer(deps = {}) {
       canvas.setAgentStatus("idle");
       open(url);
     }
-    return ok(`Canvas running at ${url}. The graph live-updates as tfstate changes. Run \`terraform plan -out=tfplan && terraform show -json tfplan > .stackcanvas/plan.json\` (or \`tofu plan \u2026\` with OpenTofu) to show the plan diff, then call await_canvas_intent to receive user edits.`);
+    const stale = canvas.getStale();
+    const warning = stale === null ? "" : `
+
+WARNING: the graph may be incomplete \u2014 ${stale}. Tell the user.`;
+    return ok(`Canvas running at ${url}. The graph live-updates as tfstate changes. Run \`terraform plan -out=tfplan && terraform show -json tfplan > .stackcanvas/plan.json\` (or \`tofu plan \u2026\` with OpenTofu) to show the plan diff, then call await_canvas_intent to receive user edits.` + warning);
   });
   mcp.registerTool("load_plan", {
     description: "Register a Terraform plan for diff highlighting on the canvas. Accepts a JSON plan (terraform show -json output) or a binary plan file.",
@@ -988,10 +1000,14 @@ function createMcpServer(deps = {}) {
     inputSchema: {}
   }, async () => {
     if (!canvas) return fail("No canvas open. Call open_canvas first.");
-    return ok(summarizeGraph(canvas.getGraph()));
+    const stale = canvas.getStale();
+    const warning = stale === null ? "" : `
+
+WARNING: the graph may be incomplete \u2014 ${stale}. Tell the user.`;
+    return ok(summarizeGraph(canvas.getGraph()) + warning);
   });
   mcp.registerTool("await_canvas_intent", {
-    description: "Block until the user clicks Apply on the canvas, then return their requested changes as intent JSON: {intent: {add, modify, remove} | null}. null = timeout, call again in a loop to keep waiting. After receiving an intent, write the Terraform code for it.",
+    description: "Block until the user clicks Send-to-agent on the canvas, then return their requested changes as intent JSON: {intent: {add, modify, remove} | null}. null = timeout, call again in a loop to keep waiting. After receiving an intent, write the Terraform code for it. After implementing an intent, run terraform plan (write the JSON to .stackcanvas/plan.json) and call this tool again to keep the loop alive. Never run terraform apply unless the human explicitly asks.",
     inputSchema: {
       timeoutSeconds: z2.number().positive().max(3600).default(45).describe("How long to wait before returning {intent: null}. Default 45s: MCP clients commonly abort requests after 60s, so keep this under your client's request timeout and call the tool in a loop instead of passing large values.")
     }

package/package.json

@@ -1,8 +1,8 @@
 {
   "name": "stackcanvas",
-  "version": "0.2.0",
+  "version": "0.2.2",
   "type": "module",
-  "main": "src/server.ts",
+  "main": "dist/cli.js",
   "bin": {
     "stackcanvas": "dist/cli.js"
   },
@@ -28,5 +28,24 @@
     "@stackcanvas/server": "workspace:*",
     "@types/node": "^22.0.0",
     "tsup": "^8.0.0"
-  }
+  },
+  "description": "Live infrastructure canvas for Claude Code: see what your AI agent is about to do to your Terraform \u2014 and approve it \u2014 before it happens. Local-first, credentials never leave your machine.",
+  "license": "MIT",
+  "repository": {
+    "type": "git",
+    "url": "git+https://github.com/pshenok/stackcanvas.git"
+  },
+  "homepage": "https://stackcanvas.dev",
+  "bugs": "https://github.com/pshenok/stackcanvas/issues",
+  "author": "Kostyantyn Pshenychnyy",
+  "keywords": [
+    "terraform",
+    "opentofu",
+    "mcp",
+    "claude-code",
+    "infrastructure-as-code",
+    "devtools",
+    "visualization",
+    "ai-agent"
+  ]
 }