aem-eds-personalization
Section-based personalization for AEM Edge Delivery Services pages authored in DA.
Authors personalize a section with Personalization rows in its Section Metadata. The section's
authored content is the default (what bots, failures, timeouts and visitors without consent see);
variants are EDS fragments chosen by rules, a decision engine, an A/B split or Adobe Target.
Consent comes first (no consent, no personalization; exceptions only with privacy/legal
approval). Each decision has one outcome and its own pzn:* measurement events, and the browser
trusts the edge's decisions only when consent allows them.
| Section Metadata | |
| Personalization | home-hero |
| Personalization Source | rules (default) | api | split | target |
| Personalization Variants | /fragments/personalization/home-hero/india |
| | /fragments/personalization/home-hero/na-mobile |
| Personalization Rule India | geo: IN |
| Personalization Rule NA Mobile | geo: US, CA & device: mobile |
| Source | Decides |
|---|---|
rules | First matching Personalization Rule <Variant> row: geo, device, new/returning visitor, URL/UTM param, page state, custom audiences |
split | A/B(/n) test against the authored content (the control), weighted by Personalization Split, kept for the session |
api | A decision engine over the v1 API (OF1 through the reference adapter), the Rule rows as the written fallback |
target | Adobe Target (Delivery API or Web SDK) with JSON { "fragment" } / { "control": true } offers; inline, or a popup on scroll/view/load |
Install
/plugin marketplace add adobe/skills
/plugin install aem-eds-personalization@adobe-skills
Then, in an EDS project:
/aem-eds-personalization:personalize geo hero on /home: India, US, Europe; TODO copy
/aem-eds-personalization:personalize A/B test the closing CTA section on /home 50/50
Skills
| Skill | Description |
|---|---|
personalize | Installs the client runtime and scripts.js hooks, records and wires consent, authors Section Metadata and variant fragments, validates (JSON Schemas, simulator, project lint, Playwright checks for outcomes, events, re-evaluation, consent changes, faults, splits, popups, page speed and the pilot scenarios) and delivers to DA only on confirmation: the push, the DA upload and preview, and the publish each ask first, and publishing waits until no copy is TODO: and main serves this runtime. Optional Cloudflare edge worker that decides rules, split and api sections before the HTML leaves the CDN (applied / selected / left, LCP-safe) and hosts the engine's server route. |
Scope
- DA projects only; xwalk and document-based projects stop at preflight.
- Section Metadata must reach the page as section attributes: the pipeline's rendering v2, or an
aem.jsthat still reads Section Metadata client-side. - Not for creating Target activities, audiences or offers, AJO journeys, or text-token replacement (placeholders sheet). Page-level experiments stay with AEM Experimentation (a page running one gets no personalized sections).
- No personal data is read, stored or sent; engine keys stay on the server (a Worker secret).
- Works on any EDS site, and on top of a migration (for example after the
stardustplugin'sdeploy) when the use case needs it. Nothing in another plugin chains it. - The edge worker is generated and tested locally;
wrangler deployand secrets are always handed off to the site owner, never run by the agent.
Requirements
- Node 22+,
@adobe/aem-cli16.21+ (aem up) for local preview - Playwright with Chromium in the skill folder or
~/.cache/aem-eds-personalization/playwrightforverify-preview(never installed into the site) wrangleronly for edge mode- A
DA_TOKEN(see theda-authskill inaem-edge-delivery-services) or a DA MCP server for delivery
Layout
skills/personalize/
├── SKILL.md workflow (preflight → choices → install → author → configure → validate → edge → deliver)
├── reference/ authoring, criteria, consent, decisions and measurement, client and edge modes,
│ engine API, split, Target, testing, DA
├── assets/ JSON Schemas and pilot scenarios, content templates, client runtime,
│ popup styles, Cloudflare worker (with the OF1 adapter)
└── scripts/ detect, install, validate, simulate, verify, mock API/Target (+ tests)
evals/personalize/ task.md and criteria.json (seven runs on a DA sandbox)
Tests
cd plugins/aem/edge-delivery-services-personalization/skills/personalize && npm test
The browser runtime tests run when Playwright resolves from the skill folder or the cache, and are left out otherwise.