[Doc] Add agent instructions for consistent AI-powered PR reviews

[Copilot]

Description

Adds structured instruction files to guide AI code reviewers (GitHub Copilot, Gemini, etc.) in producing consistent, high-quality reviews tailored to Mooncake's production infrastructure.

Files Added

.github/copilot-instructions.md

• Repository context: C++/CUDA serving platform, RDMA/TCP/NVMe-of protocols
• Build/test commands, pre-commit hooks, CI workflows
• Performance considerations: CUDA memory/sync, RDMA bandwidth, concurrency patterns
• PR classification and RFC process

.github/instructions/pr-review.instructions.md

• Review responsibilities: correctness, concurrency, performance (CUDA/RDMA-specific), security, build/CI, testing
• Anti-hallucination: explicit "ask, don't assume" guidance with examples
• Structured output template with severity levels (Blocker/Major/Minor/Info)
• Dedicated "Questions for Author" section
• Verification commands (build, test, pre-commit) with README/CI references
• Mooncake-specific guidance: Transfer Engine, CUDA kernels, Store consistency, integration contracts

Design

Severity-based review format enforces file:line references and actionable feedback. Anti-hallucination guidelines prevent speculative comments. Verification section provides concrete commands to validate changes.

Type of Change

• Types
  • Bug fix
  • New feature
    • Transfer Engine
    • Mooncake Store
    • Mooncake EP
    • Integration
    • P2P Store
    • Python Wheel
  • Breaking change
  • CI/CD
  • Documentation update
  • Other

How Has This Been Tested?

Documentation-only change. Files verified for correct placement and markdown formatting.

Checklist

• I have performed a self-review of my own code.
• I have formatted my own code using ./scripts/code_format.sh before submitting.
• I have updated the documentation.
• I have added tests to prove my changes are effective.

Original prompt

Add repository-level agent skills/instructions for PR review to help LLM reviewers (Copilot/Gemini/etc.) produce consistent, actionable code reviews.

Repository: XucSh/Mooncake

Requirements:

1. Add exactly two instruction files (keep it simple):

   • .github/copilot-instructions.md (general repository instructions)
   • .github/instructions/pr-review.instructions.md (PR review–specific instructions)

2. Language: English.

3. Content goals:

   • Tailored to Mooncake being a C++-heavy serving platform with CUDA/Python/CMake/Shell.
   • Provide clear "agent skills" / responsibilities for PR review: correctness, concurrency, performance (including CUDA sync/launch/memory), security & privacy, build/CI, tests.
   • Enforce a structured review output template with severity levels (Blocker/Major/Minor/Info) and a "Questions for author" section when context is missing.
   • Include guidance to avoid hallucination: if information is missing, ask; do not assume.
   • Include practical verification guidance (build/test/benchmark commands as placeholders if unknown, e.g. "consult README/CI"), and require referencing file/line when possible.

4. Ensure the files are placed in the correct paths and are well-formatted Markdown.

5. Open a pull request with an appropriate title and description summarizing the change.

This pull request was created from Copilot chat.

---

💡 You can make Copilot smarter by setting up custom instructions, customizing its development environment and configuring Model Context Protocol (MCP) servers. Learn more Copilot coding agent tips in the docs.

[XucSh]

@copilot

[coderabbitai]

Important

Review skipped

Bot user detected.

To trigger a single review, invoke the @coderabbitai review command.

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

• 🔍 Trigger a full review

---

_{Comment @coderabbitai help to get the list of available commands and usage tips.}