forgeax-engine-wechat

Use Tencent's installed wechatide-skill for the exact official CLI contract. This skill supplies Engine integration, reusable tools and field-tested diagnosis. Start with the relevant row; the historical observations in Sources are scoped results, not a requirement to repeat every past experiment.

Choose the next action

Keep the actual game authoritative

Confirm compileType: "game", the project path on the DevTools machine, AppID and intended target. Reuse supplied values. Read the installed official skill version, check its status, and resolve actual login/client setup through its UI. For a QR that needs the user, show the real image and send an explicit input notification; a progress image alone can be missed.

Build through the owning Engine/DevKit producer. Preserve the complete matched WASM/glue/provenance and artifact inventory. A successful import, preview upload, fixture or standalone GL fence cannot substitute for the ordinary App/World/Catalog path and completed Renderer scene frames.

Choose the cheapest environment that reproduces the failure. A Mac browser using hardware WebGL can accelerate owner fixes; an official simulator checks its native Host, and a phone checks its own runtime/driver. A phone-only failure needs direct phone evidence without an unchanged simulator detour. Record the actual target and backend on every result. Honor the user's current priority.

Default iteration is an owner regression, then the complete ordinary game in the official simulator, then the phone. Shared CPU work, Bundle validation and startup logic should be reproduced locally before another long phone cycle. Once that simulator path is healthy, use the phone for native download/network, device GPU, touch, lifecycle and final FPS evidence; simulator throughput cannot qualify a phone's budget. If a failure is already proven phone-only, retain that simulator control and diagnose the failing target directly.

Rendering failures without a named startup/asset cause follow forgeax-engine-rhi-debug: capture the failing frame and inspect its work/resources before speculative shader changes. For performance, first establish host headroom and a repeatable workload; separate correctness capture, intrusive profiling and uninstrumented acceptance.

Keep one owned simulator active and close it when unused. Preserve other sessions' projects and phone activity. Observe foreground before native input and stop when the target changes. Match input coordinates to a reviewed current screenshot.

Turn reproduced owner failures into narrow regressions; reuse unchanged builds and assets. Keep commands, source identity, actual interactions, completed frames, original failures and reviewed screenshots together. Publish only sanitized results with accessible image links. Account state, AppIDs, device identities, QR images and raw logs remain outside tracked sources.

Extend the owning diagnostic entrypoint when evidence is missing. Keep collection, bounded transport, cleanup, tests and interpretation together in this directory; graduate a successful temporary probe into that path before relying on it again. Do not make routine investigations depend on generated-chunk edits or private globals. Keep unavailable capabilities and incomplete reports visible.