docs(cli): document audit --offline flag

cli/audit.mdx

@@ -35,6 +35,7 @@ squirrel audit <url> [options]
 | `--visibility` | | Visibility: `public`, `unlisted`, `private` | `public` |
 | `--yes` | `-y` | Skip confirmation prompts (e.g. cloud credit spend) | `false` |
 | `--render` | | Render pages with a cloud browser for this run ([uses credits](/guides/browser-rendering); requires login) | `false` |
+| `--offline` | | Run fully offline: skip cloud features, publishing, and telemetry | `false` |
 
 ### Coverage Modes
 
@@ -113,6 +114,19 @@ squirrel audit https://example.com --publish --visibility unlisted
 Publishing requires authentication. Run `squirrel auth login` first.
 </Note>
 
+### Fully Offline Audit
+
+Run an audit with zero network calls beyond the site being audited — no cloud
+analysis, no publishing, no update checks, no telemetry. Results are stored in
+the local database as usual:
+
+```bash
+squirrel audit https://example.com --offline
+```
+
+`--offline` cannot be combined with `--publish` or `--render` (both require
+cloud access and login).
+
 ### Verbose Output
 
 ```bash

configuration/cloud.mdx

@@ -16,6 +16,7 @@ max_credits_per_audit = 200   # hard spend cap per audit; 0 = unlimited
 confirm_threshold = 50        # estimates above this prompt for confirmation
 batch_size = 20               # pages per cloud service request
 rendering = "http"            # "http" (default) or "browser"
+render_concurrency = 4        # concurrent browser-render jobs (1-8)
 ```
 
 ### `enabled`
@@ -48,6 +49,12 @@ Pages per request to per-page cloud services. Lower it only if you hit request-s
 
 Crawl fetch mode. `"browser"` renders every page with a cloud headless browser at **2 credits per page** - see the [browser rendering guide](/guides/browser-rendering). The `--render` flag on `squirrel audit` enables it for a single run without changing config.
 
+### `render_concurrency`
+
+**Default:** `4` (max `8`)
+
+How many cloud browser-render jobs run at once during a rendered crawl. Render jobs run in squirrelscan's cloud, so the usual per-host politeness delay doesn't apply to job submission - but the render workers do fetch your site, hence the cap. Raise it to speed up large rendered audits; lower it if your origin is sensitive to parallel traffic.
+
 ## Related
 
 - [Cloud overview](/cloud) - what cloud features add