GHSA-cp6q-959q-f8rh: XSS

Summary

@tiptap/core's public mergeAttributes() helper uses ordinary bracket assignment on keys returned by Object.entries(). An own proto key from JSON therefore invokes the legacy prototype setter on the fresh merged object. The function returns an object whose prototype is attacker-controlled, while Object.keys() and ordinary own-property checks show no attacker attributes.

When that result is used as a ProseMirror DOMOutputSpec attribute object, prosemirror-model's DOMSerializer.renderSpec() enumerates it with for...in and applies inherited values with setAttribute(). In a browser proof, inherited src and onerror values were copied to an <img> and the error handler executed once. This is per-object prototype manipulation; the proof does not modify global Object.prototype.

Root cause

The affected loop is conceptually:

ts const mergedAttributes = { ...items } for (const [key, value] of Object.entries(item)) { const exists = mergedAttributes[key] // ... mergedAttributes[key] = value }

Object.entries(JSON.parse('{"proto": {...}}')) includes proto. Reading mergedAttributes['proto'] resolves the inherited Object.prototype; assigning to the same key invokes Object.prototype.proto's setter and replaces mergedAttributes' prototype.

Browser reproduction

The following shape was tested with exact @tiptap/core 3.29.2 and prosemirror-model 1.25.11:

js const input = JSON.parse({ "proto": { "data-inherited-canary": "present", "src": "x-invalid://canary", "onerror": "globalThis.tiptapXss += 1" } })

const attrs = mergeAttributes(input) // Object.keys(attrs) === [] // Object.getPrototypeOf(attrs) === input.proto

const schema = new Schema({ nodes: { doc: { content: 'image' }, image: { toDOM: () => ['img', attrs] }, text: {}, }, }) const doc = schema.node('doc', null, [schema.node('image')]) const fragment = DOMSerializer.fromSchema(schema).serializeFragment(doc.content) document.body.append(fragment)

Chromium produced an image with data-inherited-canary, src, and onerror; the handler executed exactly once. Object.prototype remained clean.

Impact and preconditions

Applications that merge untrusted imported document, plugin, CMS, API, tenant, or AI-derived attribute objects can receive a prototype-manipulated result. Consumers that enumerate inherited keys, including ProseMirror's DOM serializer, can turn the hidden properties into DOM attributes and execute JavaScript in the application's origin. Own-key validation, object spread, JSON serialization, and logging can miss the inherited values. Other component consumers can read inherited authorization or configuration fields.

Tiptap's standard fixed ProseMirror schemas discard unknown document attributes, so arbitrary Tiptap JSON is not automatically exploitable in every application. A vulnerable application needs an untrusted object boundary into mergeAttributes() or a dynamic/custom extension or schema that preserves the relevant attribute object.

Affected versions

The unsafe assignment was introduced in commit ecadf7ea0a7f8f39a8496a60edf0ac8f379e6eb3 and is present in the first package tag @tiptap/core@2.0.0-alpha.0, v2.0.0, v2.27.1, v3.0.0, and current v3.29.2 source. No fixed release was found.

Recommended remediation

Reject proto before reading or assigning the key, or define copied keys as own data properties without invoking legacy setters. A minimal hardening is to skip key === 'proto'. Add regression tests using an own JSON-origin proto key and assert that the result keeps Object.prototype as its prototype, exposes no inherited attacker keys, and cannot create an event-handler attribute through DOMSerializer.

This was found during authorized dependency review and is being reported privately. No public zero-day issue has been opened.