---
title: "Documentation — PageWeave"
---

> For the full documentation index, see [PageWeave Documentation](https://pageweave.dev/docs) or fetch [llms.txt](https://pageweave.dev/llms.txt).

# Security

PageWeave's security architecture ensures isolation between platform code and user-generated content.

## Domain Isolation

| Domain | Purpose | Trust Level |
|--------|---------|-------------|
| `pageweave.dev` | Dashboard, API, MCP endpoint | Trusted platform code. Strict CSP. |
| `*.pageweave.site` | Hosted websites | Untrusted user HTML/JS. No platform cookies. |
| `assets.pageweave.dev` | Platform CSS/JS/fonts | Trusted static assets via CDN. |

Same-origin policy prevents `*.pageweave.site` JS from accessing `pageweave.dev` cookies or DOM.

## Cookie Policy

- Platform cookies set on `pageweave.dev` only
- `__Host-` prefix + `Secure` + `SameSite=Strict` + `Path=/`
- No platform cookies on `.pageweave.site` parent domain

## Content Safety

### Abuse Scoring

Every published page is scored for abuse risk:

- Obfuscation techniques (eval, base64, hidden iframes)
- Content length anomalies
- Excessive external links
- Hidden content (display:none, zero-size elements)
- Spam keywords
- Suspicious links (phishing patterns)

### Draft Pages

Unpublished pages include `<meta name="robots" content="noindex,nofollow">` to prevent premature indexing.

## Form Spam Protection

Every form is protected by ALTCHA proof-of-work. Three tiers, settable per form version or site-wide (form setting wins):

- **standard** — invisible proof-of-work (default)
- **code** — adds an accessible image/audio code challenge
- **strict** — harder proof-of-work plus the code challenge

An adaptive shield watches failed submissions per form: under sustained attack it raises protection automatically — site-wide when several forms are hit — and returns to the configured tier once pressure subsides. Owners are notified when protection escalates or is restored. Challenge and verification endpoints are rate-limited.

### Spam Scoring and the Spam Folder

Proof-of-work stops volume, not content — so every submission also passes a content check before it is stored. Six quiet signals score it: honeypot, submission speed, duplicate text, links, known scam phrases, and repeated characters. The score routes the submission:

- **Inbox** — delivered as usual: email notification, webhook, counted.
- **Spam folder** — kept in the form's spam tab, invisible in the inbox. No email notification, not counted — but the webhook still fires, with a `spam` block describing the verdict.
- **Discarded** — same silent success screen as every submission (senders learn nothing), but no webhook at all. Purged after 30 days; spam-folder rows after 90.

Nothing is deleted silently. You decide per submission: **Release** moves it back to the inbox and delivers the missed notification (the evidence — score and signals — is kept for your records); **Mark as spam** files a delivered one away.

One honest edge: an identical resubmission from the same visitor within 10 minutes counts as a repeat signal. A genuine "thanks again" can land in the spam folder — release it and the notification arrives.

## Password Protection

Environments and release permalinks can be gated behind a shared password: humans get a localized gate page and an encrypted host-only cookie (30 days, invalidated immediately on password rotation — the cookie binds a sub-second stamp), automation passes the password as a bearer header. Passwords are stored as bcrypt digests only — never in workflow state or tool-call logs. Gated surfaces always serve `Cache-Control: private, no-store`, including granted requests and hash-marked immutable URLs. The gate POST is rate-limited. See [Password Protection](/docs/environment-protection) for the full mechanics.

## Dashboard Protection

- `X-Frame-Options: DENY` on all `pageweave.dev` routes
- `Content-Security-Policy: frame-ancestors 'none'`
- Dashboard previews render user HTML in sandboxed iframes
- CORS: API never allows `*.pageweave.site` origins

## Request Handling

- Unknown Host headers → 404
- Page path sanitization: whitelist `[a-z0-9\-_/]` only
- Metadata escaping in layout templates
- Rate limiting on MCP tools per account

## API Key Security

- Keys bypass OAuth scope checks (full access)
- Key rotation from dashboard
- Audit log: all MCP tool calls recorded