cpyany 0.2.5 → 0.2.6

package/README.md

@@ -1,6 +1,7 @@
 # cpyany
 
-One-command installer for the PingHumans MCP server.
+One-command installer for cpyany, an MCP for coding agents that copy websites,
+pages, and individual elements from reference sites.
 
 ```bash
 npx cpyany setup
@@ -8,25 +9,25 @@ npx cpyany setup
 
 Auto-detects every supported AI client installed on your machine (Claude Code,
 Claude Desktop, Cursor), opens a browser to authorize, and writes the resulting
-bearer token into each client's MCP config from a single OAuth flow. Restart
-the affected clients to load `ping_humans` and `get_ping`.
+bearer token into each client's MCP config from a single OAuth flow.
+
+After setup, ask your coding agent to copy from a reference:
+
+```text
+Use cpyany to copy the hero section from https://example.com into this app.
+```
 
 ## Install paths
 
 | Client | MCP config we patch | Rule + skill we install | Auto-detected via |
 |---|---|---|---|
-| Claude Code | runs `claude mcp add --scope user --transport http …` (writes `~/.claude.json`) | `~/.claude/rules/pinghumans.md` + `~/.claude/skills/pinghumans-qa/SKILL.md` | `claude` on `PATH` |
+| Claude Code | runs `claude mcp add --scope user --transport http …` (writes `~/.claude.json`) | `~/.claude/rules/cpyany.md` + `~/.claude/skills/cpyany/SKILL.md` | `claude` on `PATH` |
 | Claude Desktop | `~/Library/Application Support/Claude/claude_desktop_config.json` (mac) / `%APPDATA%\Claude\claude_desktop_config.json` (win) — bridged via `mcp-remote` | _(none — Claude Desktop has no rules/skills mechanism)_ | the per-platform Claude config dir exists |
-| Cursor | `~/.cursor/mcp.json` | `~/.cursor/rules/pinghumans.mdc` (`alwaysApply: true`) + `~/.cursor/skills/pinghumans-qa/SKILL.md` | `~/.cursor` exists |
+| Cursor | `~/.cursor/mcp.json` | `~/.cursor/rules/cpyany.mdc` (`alwaysApply: true`) + `~/.cursor/skills/cpyany/SKILL.md` | `~/.cursor` exists |
 
-Guidance is two-tier (the same layout Context7 uses): a short always-loaded
-rule telling the agent _when_ to reach for PingHumans, and a `pinghumans-qa`
-skill with the full QA workflow that the agent loads only when it actually
-files or collects a human test. Installs that predate 0.1.0 (which appended a
-marker block to `~/.claude/CLAUDE.md`) are migrated automatically on any CLI
-invocation. `cpyany remove` cleans everything up. The published package
-also keeps `copyanything` and `pinghumans` as legacy binary aliases during
-the rename.
+Guidance is two-tier: a short always-loaded rule telling the agent when to use
+cpyany, and a deeper `cpyany` skill with the copy-from-reference workflow.
+`cpyany remove` cleans everything up.
 
 ## Restrict to one client
 
@@ -39,8 +40,7 @@ npx cpyany setup --client claude-code
 
 ## Skip the browser
 
-If you'd rather paste the snippet manually, sign in at
-[pinghumans.com/dashboard](https://pinghumans.com/dashboard) and copy the
+If you'd rather paste the snippet manually, open the app dashboard and copy the
 config block from the **MCP integration** section.
 
 ## Remove
@@ -52,5 +52,5 @@ npx cpyany remove --client claude-desktop   # just one
 
 ## License
 
-UNLICENSED — the CLI is published for ease of installation, but the source
-of the PingHumans server is not open-licensed.
+UNLICENSED — the CLI is published for ease of installation, but the service
+source is not open-licensed.

package/index.mjs

@@ -1,5 +1,5 @@
 #!/usr/bin/env node
-// pinghumans — install the PingHumans MCP server in your AI client.
+// cpyany — install the cpyany MCP server in your AI client.
 // One command: shows a one-time code, you approve it in the browser
 // (RFC 8628 device flow), your client configs get patched, you're done.
 
@@ -11,7 +11,7 @@ import { execFile } from "node:child_process";
 import { promisify } from "node:util";
 import open from "open";
 
-const VERSION = "0.2.5";
+const VERSION = "0.2.6";
 const execFileP = promisify(execFile);
 const APP_URL = process.env.PINGHUMANS_APP_URL ?? "https://pinghumans.com";
 // Hoisted with the other top-of-module consts — the entry try-block runs
@@ -87,82 +87,67 @@ const RULES_MARKER_START =
 const RULES_MARKER_END = "<!-- pinghumans-mcp:end -->";
 
 // Two-tier guidance (Context7-style):
-//   RULE_BODY  — short, always-loaded: WHEN to reach for PingHumans.
+//   RULE_BODY  — short, always-loaded: WHEN to reach for cpyany.
 //   SKILL_BODY — full workflow, loaded on demand when the agent does QA.
-const RULE_BODY = `Use the PingHumans MCP to put a real human in the verification loop whenever you build or change something you can't verify yourself — rendered UI, animation smoothness, real-device behavior, real third-party side-effects (OAuth round-trips, receipt emails, payments), or judgment calls about taste, clarity, and confusion. Even when the code looks right: you can't see the page, a human can. Don't caveat with "I'm guessing" — file a test or a poll instead.
+const RULE_BODY = `Use cpyany when the user asks you to copy, clone, recreate, or match a website, page, section, component, or visual element from a reference URL, screenshot, or existing site.
 
-Do not use for: anything verifiable with code, tests, or a local browser; objective questions answerable from docs or search.
+Do not use for: unrelated implementation work, backend-only tasks, or generic UI work with no reference to copy from.
 
 ## Steps
 
-1. Built/changed a UI? Deploy or tunnel it to a PUBLICLY reachable URL (localhost won't work — a remote human opens it), make sure \`<script src="https://pinghumans.com/qa-toolbar.js"></script>\` is in the preview build's HTML, then call \`request_human_test\` with structured \`steps\` — machine-verifiable \`check\` rules plus inline \`options\` (e.g. \`["Smooth","Janky"]\`) for qualitative steps
-2. Quick taste/preference read with no build to test? Call \`ping_humans\` (blocks ~50s)
-3. File tests early and keep working while the human tests — collect results with \`get_test_results\` at checkpoints, \`wait_for_results\` when blocked, or \`npx cpyany wait <ping_id>\` as a background task (foreground in one-shot runs)
-4. When results arrive, treat the human's verdict/comments as acceptance criteria. Use browser/Chrome checks to implement and confirm those criteria, not to overrule the worker.
-5. For the full QA workflow (verifiable steps, evidence, waiting modes, acting on results), use the pinghumans-qa skill
+1. Identify the reference URL, screenshot, or specific element/section the user wants copied.
+2. Preserve the target app's framework and style conventions; copy the look and behavior, not the reference site's codebase wholesale.
+3. Match layout, spacing, typography, color, imagery, responsive behavior, and interaction states as closely as practical.
+4. Verify against the reference with browser inspection or screenshots before reporting completion.
+5. For the full copy-from-reference workflow, use the cpyany skill.
 `;
 
 const SKILL_BODY = `---
-name: pinghumans-qa
-description: This skill should be used when the user has built or changed a UI and wants it verified by a real human, asks for human QA / human testing / human feedback on a build, or wants to check the results of a previously filed human test. Also activates for subjective gut-checks (taste, copy, design preference) via PingHumans.
+name: cpyany
+description: Use this skill when the user asks to copy, clone, recreate, or match a website, landing page, section, component, or visual element from a reference site, URL, screenshot, or design inspiration.
 ---
 
-When you've built or changed something you can't verify yourself, use PingHumans to have a real human test it and return structured results — a verdict, pinned component comments with CSS selectors, a per-step proof-of-work report, and screenshots you can open.
+cpyany helps coding agents recreate a reference site's visual design in the current codebase. The goal is to copy the user's requested look, layout, and interaction in one pass while still fitting the target app's framework and component conventions.
 
 ## When to Use This Skill
 
-- You shipped a UI change and can't see the rendered result ("does this look right on a real phone?", "is this animation smooth or janky?")
-- The check needs real-world reach a sandbox doesn't have ("sign in with Google — did it actually log you in?", "did the test purchase email a receipt?")
-- Sense-making ("complete the checkout — anything confusing or broken?")
-- A previously filed test needs its results collected
-- A quick subjective read with no build to test ("which logo looks more professional?") — use \`ping_humans\` for these
+- "Copy this homepage/hero/pricing card/nav from <url>"
+- "Make this section look like the reference"
+- "Clone this element into my app"
+- "Use this screenshot/site as visual inspiration and build it here"
+- "Match the spacing, typography, and styling of this page"
 
-## How to File a Test
+## Workflow
 
-### Step 1: Make the build reachable and instrumented
+### Step 1: Understand the copy target
 
-The \`url\` must be PUBLICLY reachable — a remote human opens it, so localhost won't work: tunnel it first (ngrok, cloudflared) or deploy a preview. The toolbar script is REQUIRED for pinned comments and step auto-verification — one line, once per project, in the preview build's HTML (root layout / index template), preview/dev only:
+Clarify whether the user wants a whole page, a section, a single component, or only a style direction. If they give a URL, inspect the exact target. If they give a screenshot, use it as the visual source of truth.
 
-\`\`\`html
-<script src="https://pinghumans.com/qa-toolbar.js"></script>
-\`\`\`
+### Step 2: Recreate in the target codebase
 
-No per-task id needed — the tester's claim link carries the task token. \`request_human_test\`'s result reports whether the script was detected (\`toolbar_detected\`) — if it warns, fix the build and redeploy before testers arrive.
+Use existing project components, styling systems, tokens, and routing patterns. Prefer copying the visual outcome rather than importing unrelated abstractions from the reference. Preserve accessibility and responsive behavior.
 
-### Step 2: File with verifiable steps
+### Step 3: Match the visible details
 
-Call \`request_human_test\` with a \`url\`, structured \`steps\`, and optional \`verdict_options\` / \`require_evidence\`. You wrote the code, so you know what "done" looks like — attach a \`check\` rule to every step you can:
+Pay close attention to:
 
-- \`{type:"url", pattern:"/pricing"}\` — tester actually navigated there
-- \`{type:"click", selector:".checkout button"}\` — tester actually clicked it
-- \`{type:"fill", selector:"input[type=email]"}\` — tester actually typed
-- \`{type:"comment"}\` — tester pinned at least one comment
+- spacing and alignment
+- typography scale and weight
+- color, gradients, borders, shadows, and blur
+- imagery, icon treatment, and decorative elements
+- hover/focus states and animation feel
+- desktop and mobile layout
 
-The toolbar observes the page and auto-verifies them; results report each step as ✓ auto (machine-verified), ✓ manual, or ✗ not done — your proof the tester did the work. Leave \`check\` null only for pure-judgment steps.
+### Step 4: Verify
 
-When a step asks a question rather than performs an action, add \`options\` (2–4 short labels, e.g. \`{text:"Scroll the homepage — smooth or janky?", options:["Smooth","Janky"]}\`): the tester answers with one tap right on the step and the pick comes back in \`steps_result[].answer\` — far richer than a bare checkmark. A step with a \`selector\` in its check also gets highlighted on the page for the tester, so prefer real selectors.
-
-### Step 3: Wait the right way
-
-The call does NOT block (QA takes minutes) — it returns \`status='pending'\` immediately. In order of preference:
-
-1. **File early, keep working.** The human tests in parallel — their time is free to you.
-2. **Interactive session with background shell tasks:** launch \`npx cpyany wait <ping_id>\` in the background — it exits the moment results land, waking you with the results.
-3. **One-shot run (subagent / final answer due this run):** do NOT end your run while results are pending — nothing can wake you afterwards, and a backgrounded waiter dies with you. Run \`npx cpyany wait <ping_id>\` in the FOREGROUND, or call \`wait_for_results(ping_id)\` in a loop until complete.
-4. **Checkpoint checks:** \`get_test_results(ping_id)\` at natural pauses.
-5. **Genuinely blocked:** \`wait_for_results(ping_id)\` blocks server-side (~45s per call, free) and returns early on news. Never hand-roll sleep loops.
-
-### Step 4: Act on the results
-
-Each result has a verdict, notes, pinned comments, per-step truth, and a screenshot URL — open and look at it. Comments carry CSS selectors that point at the exact component to change. Treat the human's verdict/comments as acceptance criteria, not suggestions. Browser, Chrome, screenshots, source checks, and getComputedStyle are implementation aids: use them to satisfy and confirm the worker's requested changes, never to decide the worker was wrong.
-
-**Clone-compare tasks** (filed with both \`url\` = the original and \`draft_url\` = your clone): before you touch the clone, call \`check_og_source(ping_id)\`. It reads the ORIGINAL's authored HTML source for exactly the elements a human flagged and diffs each against your clone. The worker's alignment feedback is the acceptance criteria: modify HTML/CSS so every requested move/match/delete is met with reasonable visual accuracy. Allow small manual error in hand-placed move/align markers, but preserve the worker's intended direction and target. Use the returned authored classes to implement the feedback; only fall back to measuring/live inspection for pairs marked \`needs_live_inspection\`, and only to satisfy the human request, not to replace it with your own visual judgment.
+Run the app, inspect the result in a browser, and compare it to the reference. Iterate until the copied result is recognizably faithful.
 
 ## Guidelines
 
-- **Pricing**: ~$0.05 per response received; filing is free, results are charged
-- **Quick polls**: \`ping_humans\` for taste questions blocks ~50s; poll \`get_ping(ping_id)\` for late arrivals
-- **Pending isn't dead**: "a tester has claimed the task and is testing right now" means results are imminent — keep waiting
+- Do not claim pixel perfection unless you verified it.
+- Do not copy private assets or secrets from the reference.
+- Do not add heavy dependencies just to mimic a small visual effect.
+- When the reference and target app conflict, preserve the target app's architecture and recreate the look locally.
 `;
 
 // ─── Entry ────────────────────────────────────────────────────────────────
@@ -280,7 +265,7 @@ async function setup(client) {
     track("setup_complete", { client_label: t });
   }
 
-  console.log(`\n${pc.green("✔")} PingHumans setup complete\n`);
+  console.log(`\n${pc.green("✔")} cpyany setup complete\n`);
   for (const { client: c, entries } of summary) {
     console.log(`  ${pc.bold(prettyClient(c))}`);
     for (const e of entries) {
@@ -301,7 +286,7 @@ async function setup(client) {
   if (restartList) {
     console.log(`\nRestart ${restartList} to load the MCP server.`);
   }
-  console.log(`Try it: ask your agent to "use ping_humans to ask if pizza is cooler than tacos."`);
+  console.log(`Try it: ask your agent to "use cpyany to copy the hero section from a reference site."`);
 }
 
 async function remove(client) {
@@ -319,7 +304,7 @@ async function remove(client) {
   }
   for (const t of targets) {
     await unpatchConfig(t);
-    console.log(`✓ Removed pinghumans from ${prettyClient(t)} config.`);
+    console.log(`✓ Removed cpyany from ${prettyClient(t)} config.`);
     await unpatchRules(t);
     track("remove", { client_label: t });
   }
@@ -494,7 +479,7 @@ async function performDeviceLogin() {
   const linkLine = `${pc.dim("Open this link to approve:")}\n${pc.cyan(authorization.verification_uri_complete)}\n\n${pc.dim("Or visit")} ${pc.cyan(authorization.verification_uri)} ${pc.dim("and enter the code above.")}`;
   console.log(
     boxen(`${codeLine}\n\n${linkLine}`, {
-      title: "Sign in to PingHumans",
+      title: "Sign in to cpyany",
       titleAlignment: "left",
       padding: 1,
       margin: { top: 1, bottom: 1, left: 2, right: 2 },
@@ -607,7 +592,8 @@ async function patchConfig(client, token) {
     // Claude Desktop doesn't natively support Streamable HTTP MCP servers
     // (only stdio). Bridge via the community mcp-remote package, which spawns
     // a stdio server that proxies to our hosted HTTP MCP.
-    config.mcpServers.pinghumans = {
+    delete config.mcpServers.pinghumans;
+    config.mcpServers.cpyany = {
       command: "npx",
       args: [
         "-y",
@@ -619,7 +605,8 @@ async function patchConfig(client, token) {
     };
   } else {
     // Cursor + others: native streamable-HTTP works.
-    config.mcpServers.pinghumans = {
+    delete config.mcpServers.pinghumans;
+    config.mcpServers.cpyany = {
       url: `${APP_URL}/api/mcp`,
       headers: { Authorization: `Bearer ${token}` },
     };
@@ -630,18 +617,16 @@ async function patchConfig(client, token) {
 
 async function unpatchConfig(client) {
   if (client === "claude-code") {
-    try {
-      await execFileP("claude", ["mcp", "remove", "pinghumans", "--scope", "user"]);
-    } catch (err) {
-      console.error(`Couldn't run \`claude mcp remove\`: ${err.message}`);
-    }
+    await execFileP("claude", ["mcp", "remove", "cpyany", "--scope", "user"]).catch(() => {});
+    await execFileP("claude", ["mcp", "remove", "pinghumans", "--scope", "user"]).catch(() => {});
     return;
   }
   const path = configPath(client);
   try {
     const text = await readFile(path, "utf8");
     const config = JSON.parse(text);
-    if (config?.mcpServers?.pinghumans) {
+    if (config?.mcpServers?.cpyany || config?.mcpServers?.pinghumans) {
+      delete config.mcpServers.cpyany;
       delete config.mcpServers.pinghumans;
       await writeFile(path, JSON.stringify(config, null, 2) + "\n");
     }
@@ -653,7 +638,7 @@ async function unpatchConfig(client) {
 // ─── Agent rules ──────────────────────────────────────────────────────────
 //
 // In addition to MCP tool descriptions, we install a short prose "rule"
-// telling the agent WHEN to reach for ping_humans. Tool descriptions only
+// telling the agent WHEN to reach for cpyany. Tool descriptions only
 // fire during tool selection; rules sit in the agent's persistent
 // instructions and bias it toward the tool earlier in reasoning.
 //
@@ -661,7 +646,7 @@ async function unpatchConfig(client) {
 //   - Claude Code → ~/.claude/CLAUDE.md (the canonical user-memory file
 //     Claude Code auto-loads at session start). Marker tags so re-install
 //     and remove are clean even if the user hand-edits the file.
-//   - Cursor → ~/.cursor/rules/pinghumans.mdc (Cursor's auto-loaded
+//   - Cursor → ~/.cursor/rules/cpyany.mdc (Cursor's auto-loaded
 //     rules dir, with `alwaysApply: true` frontmatter).
 //   - Claude Desktop → no equivalent; skipped.
 //
@@ -686,17 +671,17 @@ function stripMarkerBlock(text) {
 // mechanism, so it gets MCP config only.
 function rulePath(client) {
   if (client === "claude-code")
-    return join(homedir(), ".claude", "rules", "pinghumans.md");
+    return join(homedir(), ".claude", "rules", "cpyany.md");
   if (client === "cursor")
-    return join(homedir(), ".cursor", "rules", "pinghumans.mdc");
+    return join(homedir(), ".cursor", "rules", "cpyany.mdc");
   return null;
 }
 
 function skillPath(client) {
   if (client === "claude-code")
-    return join(homedir(), ".claude", "skills", "pinghumans-qa", "SKILL.md");
+    return join(homedir(), ".claude", "skills", "cpyany", "SKILL.md");
   if (client === "cursor")
-    return join(homedir(), ".cursor", "skills", "pinghumans-qa", "SKILL.md");
+    return join(homedir(), ".cursor", "skills", "cpyany", "SKILL.md");
   return null;
 }
 
@@ -779,7 +764,7 @@ async function refreshStaleRules({ verbose = false } = {}) {
   }
 
   if (updated.length > 0) {
-    console.log(`↻ Refreshed PingHumans agent rules (v${VERSION}):`);
+    console.log(`↻ Refreshed cpyany agent rules (v${VERSION}):`);
     for (const p of updated) console.log(`   ${p}`);
     track("rules_refresh");
   } else if (verbose) {
@@ -812,6 +797,13 @@ async function patchClaudeCodeViaCli(token) {
   // It writes to ~/.claude.json (or similar) for us.
   try {
     // Remove any existing config first, otherwise `mcp add` errors on dupes.
+    await execFileP("claude", [
+      "mcp",
+      "remove",
+      "cpyany",
+      "--scope",
+      "user",
+    ]).catch(() => {});
     await execFileP("claude", [
       "mcp",
       "remove",
@@ -829,7 +821,7 @@ async function patchClaudeCodeViaCli(token) {
       "user",
       "--transport",
       "http",
-      "pinghumans",
+      "cpyany",
       `${APP_URL}/api/mcp`,
       "--header",
       `Authorization: Bearer ${token}`,
@@ -872,7 +864,7 @@ function prettyClient(client) {
 }
 
 function printHelp() {
-  console.log(`cpyany — install the PingHumans MCP server in your AI client.
+  console.log(`cpyany — install the cpyany MCP server in your AI client.
 
 Usage:
   cpyany setup [--client claude-code|claude-desktop|cursor]
@@ -881,7 +873,7 @@ Usage:
                        # block until a human test has results (exit 0 = news,
                        # 2 = timed out). Run it in the background after filing
                        # a test so your agent gets woken when results land.
-  cpyany whoami    # show which PingHumans account this machine's token belongs to
+  cpyany whoami    # show which account this machine's token belongs to
   cpyany rules     # refresh the installed agent rules to this version
   cpyany version
 
@@ -893,7 +885,6 @@ Setup opens your browser to ${APP_URL}/cli-auth, generates a fresh bearer
 token after you sign in, and writes it into each client's MCP config.
 
 Aliases: --claude (= --claude-desktop), --code (= --claude-code).
-Legacy binary alias: pinghumans.
 `);
 }
 
@@ -911,7 +902,7 @@ Legacy binary alias: pinghumans.
 // Hoisted as a function — the entry dispatch runs before bottom-of-module
 // consts initialize (the 0.0.4 TDZ lesson, again).
 function credsPath() {
-  return join(homedir(), ".config", "pinghumans", "credentials.json");
+  return join(homedir(), ".config", "cpyany", "credentials.json");
 }
 
 async function readCreds() {
@@ -983,7 +974,7 @@ async function validateStoredToken() {
     const warn = (s) => (pc ? pc.yellow(s) : s);
     console.error(
       "\n" +
-        warn("⚠ Your PingHumans token is no longer valid") +
+        warn("⚠ Your cpyany token is no longer valid") +
         " (revoked, or the account was deleted).\n" +
         "  Re-link this machine:  " +
         (pc ? pc.cyan("npx cpyany setup") : "npx cpyany setup") +
@@ -1007,7 +998,7 @@ async function resolveLocalToken() {
   for (const p of candidates) {
     try {
       const cfg = JSON.parse(await readFile(p, "utf8"));
-      const entry = cfg?.mcpServers?.pinghumans;
+      const entry = cfg?.mcpServers?.cpyany ?? cfg?.mcpServers?.pinghumans;
       const header = entry?.headers?.Authorization ?? entry?.headers?.authorization;
       const m = /Bearer\s+(\S+)/.exec(header ?? "");
       if (m) return m[1];
@@ -1027,7 +1018,7 @@ async function waitForResultsCli(rest) {
   const token = await resolveLocalToken();
   if (!token) {
     throw new Error(
-      "No PingHumans token found. Run `npx cpyany setup` first."
+      "No cpyany token found. Run `npx cpyany setup` first."
     );
   }
 
@@ -1061,7 +1052,7 @@ async function waitForResultsCli(rest) {
 
     if (status === "not_found") {
       throw new Error(
-        "Ping not found for this account. Results are asker-scoped — `wait` only works on tests filed with the same PingHumans account."
+        "Ping not found for this account. Results are asker-scoped — `wait` only works on tests filed with the same cpyany account."
       );
     }
     if (status !== "pending" || received > baseline) {

package/package.json

@@ -1,13 +1,11 @@
 {
   "name": "cpyany",
-  "version": "0.2.5",
-  "description": "Install the PingHumans MCP server in Claude Desktop, Claude Code, or Cursor with one command.",
+  "version": "0.2.6",
+  "description": "Install cpyany so coding agents can copy websites, pages, and elements from references.",
   "type": "module",
   "main": "index.mjs",
   "bin": {
-    "cpyany": "index.mjs",
-    "copyanything": "index.mjs",
-    "pinghumans": "index.mjs"
+    "cpyany": "index.mjs"
   },
   "files": [
     "index.mjs",
@@ -26,12 +24,6 @@
     "ora": "^9.4.0",
     "picocolors": "^1.1.1"
   },
-  "homepage": "https://pinghumans.com",
-  "repository": {
-    "type": "git",
-    "url": "git+https://github.com/alex-durango/pinghumans.git",
-    "directory": "cli"
-  },
   "license": "UNLICENSED",
   "private": false,
   "publishConfig": {
@@ -43,9 +35,11 @@
     "claude",
     "cursor",
     "ai",
-    "humans",
     "cpyany",
-    "copyanything",
-    "pinghumans"
+    "copy",
+    "website",
+    "design",
+    "reference",
+    "clone"
   ]
 }