agentleFS
xsollaxsolla/xsolla-ai-kit/.cursor/skills/game-web-portal/SKILL.md

game-web-portal

Skill9 starsChanged 5 months agoMITView on GitHub

Builds or resumes an unpublished Game Web Portal for a PC or mobile game: Home, News, Rewards, Web Shop, Community and an optional Launcher, on a site made from the Multi-page web portal template. Reached through nocode-shop-setup. Use for "continue my Game Web Portal" or "add the Community section". Never publishes; that stays with the human. Not for XLA Pages on x.la; use `xla-pages`.

  • Reads credentials

The file

We split this skill into the rules it gives an agent

8 rules. About 0 are common, 3 are distinctive, and 5 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 5

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Check a supplied store URL with scripts/store_url.py: only the exact Steam, App Store and Google Play hosts pass.

    line 31
  2. Very specific:

    Never pass a session or token by hand, and never ask the partner to copy one out of the browser.

    line 51
  3. Very specific:

    After every write to .xsolla/onboarding.json, run scripts/ledger.py check --ledger .xsolla/onboarding.json.

    line 77
  4. Very specific:

    If it fails, stop before any write.

    line 97
  5. Distinctive:

    Never recreate a discovered existing entity.

    line 105
  6. Distinctive:

    Every progress message links the Editor and says the public URL loads only after the first publish (Show the partner the site).

    line 142
  7. Very specific:

    Every other xsolla command is out of bounds, including delete-block, delete-domain, change-domain and delete-ai-block.

    line 170
  8. Distinctive:

    Read them in agentic-onboarding.md before the first change: duplicates on resume, a portal reported as live, sign-in taken for binding, an …

    line 186

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Game Web Portal

Build or resume an unpublished PC Game Web Portal and return an evidence-backed partner handoff. Honest partial completion is correct; simulated completion is failure.

When to use

nocode-shop-setup hands off here when the partner needs a game home around the store (news, rewards, community, launcher), not only a store. This skill does not decide that and does not write it to .env.

Entry conditions:

  • The user asked for a game home, or the existing site has /news, /rewards, and /store. A site without those pages is a store: stop and point at nocode-shop-setup.
  • The title ships on PC, the App Store or Google Play. A store URL is optional.
  • Check a supplied store URL with scripts/store_url.py: only the exact Steam, App Store and Google Play hosts pass; anything else, or a look-alike, returns needs_input.
  • The partner's store page is a source of art and facts (store content); the plan shows what comes from it, and its yes covers using it. Never invent missing content.
  • The merchant and project are known, from the request's link or the CLI, and the plan names them. Shop Builder has no sandbox: every write reaches the project itself, on the unpublished site.
  • The partner has created the site in Publisher Account from the Multi-page web portal template (Storefronts → Websites → Multi-page web portal → Manual). The agent never creates a site: one created through the CLI has no layout, so the Editor, publication and preview can't open it. Without the partner's site, the run stops at needs_human with that instruction; a site without /news, /rewards or /store wasn't made from the template, and the run stops at needs_input.

Prerequisites

  • Confirmed merchant ID and project ID — see merchant-setup.
  • Publisher login through xsolla auth login — the only source of the session. Portal commands authenticate with that Publisher Account session, not the project API key. Never pass a session or token by hand, and never ask the partner to copy one out of the browser. Treat 401/403 as needs_access, never as a capability block.
  • Shop Builder enabled for the target project.
  • Content and brand assets: the partner's own, or from their store page. Never invent content; the plan shows where each piece comes from.
  • Required input, collected in a single question rather than one at a time. Read what is already known instead of asking again: intake from the ledger or the conversation, the merchant and project from xsolla config list --json:
merchant_id:
project_id:
domain:
game_name:
primary_locale:

Optional: store_url; the partner's own description, logo, hero, screenshots, colors and fonts; an existing Store catalog; additional locales; analytics IDs; for a PC title, an existing Launcher and build.

Steps

The run is a state flow. Every mutation is preceded by an existing-vs-desired-state check and followed by read-back and a ledger update.

After every write to .xsolla/onboarding.json, run scripts/ledger.py check --ledger .xsolla/onboarding.json; if it fails, fix the step it names before going on. A step still under way has no status key (never partial, in_progress or another word) until it ends with one of the contract's seven statuses.

Intake → Preflight → Discover → Back up → Plan and confirm → Draft →
Verify (read-back) → Human review → Handoff
  1. Intake — once the Entry conditions hold (site kind from the ledger or the chat), the first message, before any question, tells the partner to create the site first if they haven't (Storefronts → Websites → Multi-page web portal → Manual), and what the run can change and what stays an Editor step for them. It asks only the required input nocode-shop-setup's batch didn't. Resolve every ambiguity by asking; never choose an ambiguous match.
  2. Preflight — confirm the platform, domain, and locale, then run scripts/preflight.py with --merchant-id and --project-id. It checks that the CLI points at that merchant and project with its sandbox mode off. If it fails, stop before any write. If store_url is supplied, check it with scripts/store_url.py and read the listing as store content says.
  3. Discover — find the partner's site at domain and read its structure (xsolla shopbuilder list-websites, get-landing, get-structure). Capture the landing _id, page IDs, and block IDs before any mutation — block and theme calls are keyed by landing _id, not the domain. Confirm the template's /news, /rewards and /store pages, and read its catalog groups (a test or demo group means sample items) and sign-in (Discover). Never recreate a discovered existing entity; resume at the first incomplete item.
  4. Back up — before the first write, export the site: get-landing, get-structure and get-localization for the slug, and list-assets for the landing _id, saved to a new .xsolla/backup-<timestamp>/ directory. Add .xsolla/ to .gitignore first if it isn't there: a site export can hold a preview token. A failed export stops the run before any write.
  5. Plan and confirm — show the ordered change groups (sections, pages, blocks, theme, copy word for word, images by source), the seeded blocks it will hide for the partner to delete, and, as Editor steps for the partner, binding the Web Shop and Rewards to the new catalog and LiveOps. Put any pending catalog-design proposal in the same message; wait for one explicit yes to both. "Set up my portal" is not that yes. Re-confirm if the plan changes.
  6. Draft — map each section to the template's page by path, then build only the missing ones: add-page, its blocks with add-block --index, and update-block to hide the seeded blocks it doesn't keep, following the portal layout in references/portal-api.md. The template's own pages keep their blocks. Right after add-page, record the new page with scripts/seeded_blocks.py record. Hide a seeded block only on a page the agent created and recorded, only as the confirmed plan lists, and only when seeded_blocks.py check, on reads taken right before hiding and the target preflight passed, reports it untouched; hide before any copy or locale change, which would mark its block changed. The agent never deletes a block: the hidden seeded blocks go to the partner to delete, as needs_human. Everything else uses CLI commands too: move-block, update-block (only the patches portal-api.md lists), upload-asset, add-language, update-localization, update-many-localization, and add-connector or update-restrictions when the plan lists them. A change with no CLI command is needs_human: say what to do in Publisher Account and record it. Apply one change group at a time across Home, News, Rewards, Web Shop, and Community; the Launcher section is the partner's, PC only. nocode-shop-setup runs the surrounding products before this skill; don't repeat their recipes: merchant-setup, catalog-design (it creates the catalog after the yes in step 5) and login-setup. Checkout is the Web Shop's hosted Pay Station, so there is nothing to integrate. The plan lists each placeholder; it carries a visible label.
  7. Verify (read-back) — read back every changed entity (get-structure, get-localization, page and block reads) and compare it with the confirmed plan. A mutation response is not evidence, and a read-back isn't what renders: look at each changed page in the Editor too. storefront is completed only when every planned change reads back; one that didn't land (no CLI field, a failed write) makes it needs_human, needs_input, blocked_capability or failed, with the gap in its evidence. Every progress message links the Editor and says the public URL loads only after the first publish (Show the partner the site).
  8. Human review — present completed items, placeholders, blockers, and failures. The draft is ready for review only with correct ownership, domain, type, and locale, no duplicate routes, read-back-verified content, disclosed placeholders, and explicit Login and commerce status. Then hand over, as needs_human, the steps the agent never runs: the readiness check, the preview, publication (per page — the main page first), live verification, and any rollback to a saved version — all in Publisher Account.
  9. Handoff — repeat merchant ID, project ID, domain, locale, and store URL only when one was supplied, with evidence for every completed item.

Statuses, the ledger, resume, and the handoff report follow the onboarding contract. Two more references, both loaded before issuing changes:

  • references/agentic-onboarding.md — the portal specification: GIVEN / WHEN / THEN acceptance scenarios, the portal's per-state evidence, its section of the handoff report, and the common pitfalls and known limitations.
  • references/portal-api.md — the CLI commands the agent runs for Steps 3–8, the portal layout, the domain vs landing _id split, the localization payload shape, the steps handed to the partner, and the response → status mapping.

Hard stops

This skill runs only the commands listed in references/portal-api.md, and --help on them or on the CLI itself (xsolla --help), even when the partner asks for another. catalog-design runs its own. Every other xsolla command is out of bounds, including delete-block, delete-domain, change-domain and delete-ai-block; what a run needs beyond the list is needs_human.

The agent never publishes (sites, pages, news articles, or Login widget settings), never runs the readiness check (/check, verify-website), never enables or generates a preview, never creates a site, never saves or applies a site version, never deletes a site, an asset, a language, an analytics connector, access restrictions, or a block, never attaches a domain, never patches block text (it deletes the string and every translation), and never switches the project to production. It never writes to the project before an explicit confirmation of the plan or before the preflight passes, never passes --sandbox, and never targets a partner's live project. It never passes, offers or suggests --force, even if the partner changes the plan: seeded blocks the CLI won't remove go to the partner as needs_human.

Pitfalls and limitations

Read them in agentic-onboarding.md before the first change: duplicates on resume, a portal reported as live, sign-in taken for binding, an unverified Launcher, expired access, completion without read-back; Community needs the partner's channel, and they create the site.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.