Skip to content

Open-source platform for 3D AI agents. Turn text or a photo into a rigged, animated GLB avatar, give it an LLM brain, memory and a wallet, and embed it anywhere with one web component. Remote MCP server, x402 payments, Three.js, browser-native.

fix(guards): register the computer-use operability sweep

nirholase76a041cc583
diff --git a/data/guards.json b/data/guards.json
index 339cf7adb..9d44bf2cf 100644
--- a/data/guards.json
+++ b/data/guards.json
@@ -40,6 +40,22 @@
}
],
"guards": [
+ {
+ "id": "audit-agent-operability",
+ "script": "scripts/audit-agent-operability.mjs",
+ "npm": "audit:agent-operability",
+ "title": "Computer-use operability sweep",
+ "protects": "The primary task on /create, /forge, /connect, an agent profile chat, and /launch stays reachable with accessibility-tree locators alone (getByRole/getByLabel), the way a screen reader or a computer-use agent drives it.",
+ "why": "Grok Bot and any other computer-use agent never clicks a CSS selector or a raw coordinate; it enumerates roles and labels. A control that only works via hover reveal or a mouse-specific gesture is invisible to that kind of agent even though every human-facing test passes, so this drives the real surfaces the same restricted way and records any step that needed something else.",
+ "stages": [
+ "manual"
+ ],
+ "needs": "browser",
+ "proof": {
+ "kind": "live",
+ "reason": "Drives five real pages with Playwright role/label locators against a running server, including a signed-in session for avatar-limited steps. A static fixture cannot prove an accessibility-tree locator resolves against live markup."
+ }
+ },
{
"id": "audit-console",
"script": "scripts/audit-console.mjs",
diff --git a/docs/guards.md b/docs/guards.md
index 53696ffd4..706ad1159 100644
--- a/docs/guards.md
+++ b/docs/guards.md
@@ -184,6 +184,7 @@ Run `npm run audit:guards` to print the current count and per-stage breakdown. T
| Live CSP sweep | `npm run audit:csp` | No page violates the Content-Security-Policy the server sends with it, and every response carries the security headers `vercel.json` declares. `--headers-only --base https://three.ws` checks the header half against a live origin with no browser. |
| Served asset length | `npm run check:asset-encoding` | The largest compressible files under `public/` arrive in full from the live site when requested with a browser's exact `Accept-Encoding`, so a CDN variant whose declared length disagrees with its body cannot hang a page. |
| Console sweep | `npm run audit:console` | A clean browser console on every route, desktop and mobile. Any route that fails in the parallel pass is re-run once serially and the second reading is the one reported, because this worktree is shared and a page that misses its settle window beside somebody else's full build has not failed. The report names how many cleared that way. |
+| Computer-use operability sweep | `npm run audit:agent-operability` | The primary task on `/create`, `/forge`, `/connect`, an agent profile chat, and `/launch` stays reachable with accessibility-tree locators alone (`getByRole`/`getByLabel`), the way a screen reader or a computer-use agent drives it. |
| Hosted IBM page | `npm run audit:ibm-hosted` | The publish-once partnership page still works from a domain that is not three.ws: the live update applies, i18n mounts, and the x402 widget binds. See [ops/ibm-hosted-page-audit.md](ops/ibm-hosted-page-audit.md). |
| `/play` failure modes | `npm run audit:play-failures` | `/play` stays usable when its dependencies fail, and hostile deep-link params never execute. |
| Overlapping fixed overlays | `npm run audit:overlays` | No persistent floating widget can cover another one's controls. |
diff --git a/docs/ops/guard-wiring.md b/docs/ops/guard-wiring.md
index 768259a82..e33f0c541 100644
--- a/docs/ops/guard-wiring.md
+++ b/docs/ops/guard-wiring.md
@@ -148,6 +148,7 @@ viewports. They are pre-event, pre-release and incident tools.
| Guard | What it checks | Exit | Runtime | Verdict |
|---|---|---|---|---|
| `audit:console` | Console errors, page errors and failed requests on every route in `data/pages.json`, desktop and mobile | capped | >600s (reached 591 of 796 route-viewport pairs in 600s) | **manual (pre-release).** |
+| `audit:agent-operability` | Drives `/create`, `/forge`, `/connect`, an agent profile chat, and `/launch` with `getByRole`/`getByLabel` locators only, the way a computer-use agent or screen reader drives a page | 0 | needs a live `BASE_URL` plus `audit:web:login` for the avatar-limited and `/launch` steps | **manual (x-grok, pre-release).** |
| `audit:web` / `audit:web:login` | The full authed page sweep: console, network, layout | capped | >600s | **manual.** See [page-audit.md](page-audit.md). `audit:web:login` mints the session `audit:web` replays, so it is the same sweep plus a login. |
| `audit:a11y` | Playwright accessibility pass over the top pages | killed at cap | >578s (44 tests, one worker, several minutes each) | **manual.** It is also a second `playwright test` run, which fights the shared dev server on :3000. |
| `audit:overlays` | That no floating widget covers another, across routes and viewports | capped | >240s | **manual.** |
diff --git a/package.json b/package.json
index 3e30f9681..df420be51 100644
--- a/package.json
+++ b/package.json
@@ -221,6 +221,7 @@
"x:archive:analyze": "node scripts/x-archive-analyze.mjs",
"x:volume": "node scripts/x-volume-study.mjs",
"audit:deploy": "node scripts/audit-deploy-artifacts.mjs",
+ "audit:agent-operability": "node scripts/audit-agent-operability.mjs",
"guard:esbuild": "node scripts/guard-esbuild-bundles.mjs",
"guard:esbuild:all": "node scripts/guard-esbuild-bundles.mjs --all",
"verify:solana": "node scripts/verify-solana-parity.mjs",
diff --git a/public/guards.json b/public/guards.json
index 339cf7adb..9d44bf2cf 100644
--- a/public/guards.json
+++ b/public/guards.json
@@ -40,6 +40,22 @@
}
],
"guards": [
+ {
+ "id": "audit-agent-operability",
+ "script": "scripts/audit-agent-operability.mjs",
+ "npm": "audit:agent-operability",
+ "title": "Computer-use operability sweep",
+ "protects": "The primary task on /create, /forge, /connect, an agent profile chat, and /launch stays reachable with accessibility-tree locators alone (getByRole/getByLabel), the way a screen reader or a computer-use agent drives it.",
+ "why": "Grok Bot and any other computer-use agent never clicks a CSS selector or a raw coordinate; it enumerates roles and labels. A control that only works via hover reveal or a mouse-specific gesture is invisible to that kind of agent even though every human-facing test passes, so this drives the real surfaces the same restricted way and records any step that needed something else.",
+ "stages": [
+ "manual"
+ ],
+ "needs": "browser",
+ "proof": {
+ "kind": "live",
+ "reason": "Drives five real pages with Playwright role/label locators against a running server, including a signed-in session for avatar-limited steps. A static fixture cannot prove an accessibility-tree locator resolves against live markup."
+ }
+ },
{
"id": "audit-console",
"script": "scripts/audit-console.mjs",