GitHub research guide: TanStack/query

Origin: https://github.com/TanStack/query
Expected value: Async server-state and mutation architecture.
Rights flag: MIT. Do not treat this as license verification for a particular file or release.

How to use this upstream project without MCP

1. Inspect the actual task and determine whether query supplies a feature not already present in the Perfect_AI pack. Do not import a whole GitHub repository merely to satisfy a trend.
2. Check the original repo's README, release history, LICENSE, install guide and installed dependency versions at the time of work.
3. Read the upstream implementation and extract only the necessary design principle or, with license compliance, the needed source component.
4. Prefer official registry or npm install commands over arbitrary curl | bash scripts. Preserve project lockfiles, make changes in a branch, review generated diffs and dependencies.
5. Adapt design to real UI contexts: 320px container, desktop, RTL Persian, LTR English, reduced motion, input from touch and keyboard, true page content and honest metrics.
6. Ensure observable behavior is tested in an actual browser. Document what was implemented and which upstream source was only consulted.

Integration brief

• Desired value: Async server-state and mutation architecture.
• Fallback: semantic HTML with CSS; no JS-only critical UI.
• Style: honor user color and typography brief, specifically strict monochrome on Perfect_AI.
• Animation owners: never let two libraries animate the same transform independently.
• Accessibility: keyboard actions, readable text, visible focus, correct labels, and state-independent semantics.
• Cleanup: remove event handlers, dispose GPU assets, stop animations on unmount, handle React Strict Mode.

Required acceptance

• Source inspected; exact upstream revision identified.
• License and any non-MIT restrictions considered.
• Only task-relevant portions integrated and attribution preserved as needed.
• Independent compile/build and browser QA completed.
• No MCP, no remote execution without human review.
• Report source and changed files, actual testing evidence and open risks.

Classification: locally authored integration guidance, not a vendored original GitHub skill or codebase.