[AI] Moved PR Reviewer Agent to File-Based Instruction System

[kubaflo]

Description

This PR migrates the .NET MAUI PR Reviewer Agent from a large inline file prompt to a file-based instruction architecture. The previous inline prompt exceeded GitHub’s token limits, preventing the agent from loading. Because the instruction set is large and cannot be shortened, externalizing it into Markdown files is the correct and scalable solution.

Changes

• Moved the full PR reviewer prompt from YAML into Markdown files.
• Added a dedicated folder structure for agent instructions:

.github/
  agents/
    pr-reviewer.md
  instructions/pr-reviewer-agent
    core-guidelines.md
    testing-guidelines.md
    safearea-guidelines.md
    sandbox-setup.md
    edge-cases.md

• Updated the agent YAML to load instructions from these files:

instructions:
  - file: .github/agents/pr-reviewer.md
  - file: .github/instructions/pr-reviewer-agent/core-guidelines.md
  - file: .github/instructions/pr-reviewer-agent/testing-guidelines.md
  - file: .github/instructions/pr-reviewer-agent/safearea-guidelines.md
  - file: .github/instructions/pr-reviewer-agent/edge-cases.md
  - file: .github/instructions/pr-reviewer-agent/sandbox-setup.md

• No functional changes to agent behavior.

Why This Change Is Needed

• Inline prompts have strict token limits.
• The PR reviewer agent contains extensive instruction logic that cannot be shortened.
• This structure improves maintainability and aligns with GitHub’s recommended agent architecture.

Impact

• No behavioral changes.
• Agent now loads reliably without token overflow issues.
• Instructions are easier to maintain, reorganize, and extend.

Validation

• Agent successfully loads with file-based instructions.
• All referenced .md files resolve correctly.
• No token limit errors.
• Manual testing confirms identical behavior to the previous version.

Before	After

[dotnet-policy-service]

Hey there @@kubaflo! Thank you so much for your PR! Someone from the team will get assigned to your PR shortly and we'll get it reviewed.

[Copilot]

Pull Request Overview

This PR addresses GitHub token limit issues by migrating the PR Reviewer Agent instructions from a large inline YAML file to a file-based instruction architecture. The change splits ~1200 lines of inline instructions into 7 organized markdown files under .github/instructions/pr-reviewer-agent/.

Key changes:

• Created modular instruction files for testing, sandbox setup, error handling, output format, edge cases, SafeArea testing, and core guidelines
• Updated main agent file to reference external instruction files
• No functional behavior changes intended

Reviewed Changes

Copilot reviewed 8 out of 8 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
.github/agents/pr-reviewer.md	Reduced from ~1200 lines to 39 lines; now references external instruction files
.github/instructions/pr-reviewer-agent/core-guidelines.md	Core review philosophy, workflow, and code analysis patterns
.github/instructions/pr-reviewer-agent/testing-guidelines.md	App selection (Sandbox vs HostApp), PR fetching, build/deploy procedures
.github/instructions/pr-reviewer-agent/sandbox-setup.md	Sandbox modification patterns and instrumentation examples
.github/instructions/pr-reviewer-agent/safearea-guidelines.md	SafeArea-specific testing guidance
.github/instructions/pr-reviewer-agent/output-format.md	Review structure and redundancy elimination guidelines
.github/instructions/pr-reviewer-agent/error-handling.md	Build error and unexpected result handling procedures
.github/instructions/pr-reviewer-agent/edge-cases.md	Edge case discovery requirements

[Copilot]

Incorrect file extension. The repository uses .sln files, not .slnx. These commands should reference Microsoft.Maui.sln instead of Microsoft.Maui.slnx.

Suggested change

	dotnet clean Microsoft.Maui.slnx
	rm -rf bin/ obj/
	dotnet restore Microsoft.Maui.slnx --force
	dotnet clean Microsoft.Maui.sln
	rm -rf bin/ obj/
	dotnet restore Microsoft.Maui.sln --force

[PureWeen]

/rebase