OCC Headless Frontend Skill

AI coding agent skill for building a headless storefront on SAP Commerce OCC REST API without Spartacus (Angular, React / Next.js, Vue / Nuxt).

It covers what a frontend developer needs on SAP Commerce Cloud 2211-jdk21:

• Login: authorization code + PKCE, custom login pages, and a BFF with httpOnly cookies.
• Carts: anonymous carts, merge at login, guest checkout and PSP plugins.
• Catalog: product and search pages.
• CMS: SAP CMS pages and components, SmartEdit, and an external CMS next to OCC.
• B2B: orgUsers calls and pay on account.
• Debugging: CORS and 401 troubleshooting.
• Framework recipes: Next.js, Nuxt and Angular, each with SSR, BFF and SPA variants.

There is also a catalogue of SAP-specific behaviour that frontend developers don't expect (references/sap-quirks.md).

Installation

npx skills add Emenowicz/occ-headless-skill

Tested

Backend. A vanilla SAP Commerce Cloud 2211-jdk21.19 (local, HSQLDB) with the electronics and powertools sample stores, commercewebservices, b2bocc, cmsocc and the SmartEdit extensions.

What was verified live. Statements marked (live) in the references were reproduced there:

• the authorization server: grants, PKCE, code reuse, refresh rotation, session lengths and the custom login page;
• OCC error formats and fields rules;
• the cart lifecycle, merge errors, concurrent writes and checkout request formats;
• CMS pages and components;
• SmartEdit preview tickets;
• the B2B orgUsers rules;
• CORS.

What was not verified live. Statements tagged [SAP Help], [OpenAPI] or [Spartacus] come from those public sources. Not checked live:

• CCv2-only behaviour: custom domains, IP filters, CDN;
• the SmartEdit UI in a browser;
• payment plugins (the Adyen roles and redirect flow are read from the plugin's public source);
• Storyblok and Contentful apps.

Benchmark. Built with Anthropic's skill-creator loop: realistic tasks in each framework, run with and without the skill, graded blind against written expectations. The eval set is in evals/, and every row below is one run per configuration (Opus 5.5).

Iteration	Evals	With skill	Without skill
2	10: cart, login, CORS, CMS, PLP, checkout, B2B, PDP, two code reviews	96%	81%
3	9: the areas the fixes touched, plus SmartEdit and Storyblok	94%	80%

The largest gains are on 2211-jdk21 login and cart merge (100% vs 50%), SAP CMS rendering and SAP-specific code review.

Other models, with the skill. The login, cart and review evals scored 97% on Sonnet 5 and 62% on Haiku 4.5. Use a Sonnet- or Opus-class model. With smaller models, check the output against the hard rules in SKILL.md.

Triggering. On 20 realistic queries (evals/trigger-evals.json), the description triggered on 7 of the 10 frontend tasks (two runs each) and never on the 10 near-misses:

• a Java OCC extension, Spartacus customization, ImpEx, HAC Groovy, a CCv2 build;
• the accelerator JSP storefront;
• Storyblok alone, Shopify, S/4 on BTP, Auth0.

The misses were short debugging questions that the model answered without loading a skill.

Caveats.

• There is one run per cell.
• A few grades were corrected after a live or source check, each documented in the grading file.
• The absolute numbers matter less than the gaps on the SAP-specific tasks.
• The current version adds corrections made after the last benchmark run, each checked live or against public source. They were not re-benchmarked.

Independence

This skill is not affiliated with or endorsed by SAP. It is built only from public documentation (help.sap.com, the OCC OpenAPI spec, the open-source Spartacus code) and a vanilla SAP Commerce distribution. It contains no customer or employer code.