feat: Integrate GitHub Copilot SDK

[SergUdo]

Summary

Integrate GitHub Copilot SDK as a native LLM provider (CopilotProvider) into the
ai-slop-gate provider pipeline. Currently a stub exists in planning only. This issue
tracks the full implementation.

Background

ai-slop-gate supports multiple LLM providers (Gemini, Groq, Ollama) via a
common LlmProvider base class. GitHub Copilot has announced an extensibility SDK
that allows programmatic access to Copilot's code-aware reasoning. This provider
would add Copilot as a zero-cost option for GitHub-native teams (no separate API key
required for users with an active Copilot subscription).

Goals

• Implement CopilotProvider extending LlmProvider with kind = "llm", name = "copilot"
• Replace the current StubCopilotClient with a real CopilotClient backed by the SDK
• Support both analyze(diff) and collect(base_path) modes
• Implement analyze_pr(repo, pr_id, token) for native GitHub PR integration
• Add prompt file at providers/llm/prompts/copilot/deep.prompt
• Register --provider copilot in CLI
• Wrap with CachedProvider (same as Gemini/Groq)

Non-Goals

• No changes to PolicyEngine, reporters, or cache layer
• No Copilot-specific enforcement logic
• No changes to existing providers

Architecture

CopilotProvider (providers/llm/copilot.py)
  extends LlmProvider
  kind = "llm"
  name = "copilot"
  
  analyze(code, input_file) -> ProviderObservation
    → loads deep.prompt
    → calls CopilotClient.run(task)
    → parses and validates JSON response
    → retry once with force_strict_json=True on InvalidAgentResponse
    → returns ProviderObservation

  analyze_pr(repo, pr_id, token) -> ProviderObservation
    → fetches PR diff via GitHub API
    → delegates to analyze(diff)

JSON contract (same as all LLM providers):

{
  "issues": [
    {
      "category": "string",
      "signal": "snake_case",
      "message": "non-empty",
      "severity": "low|medium|high|critical",
      "confidence": 0.85,
      "file": "path/to/file.py",
      "line": 42
    }
  ]
}

Implementation Steps

• Monitor Copilot SDK public availability and API stability
• Implement CopilotClient (replaces StubCopilotClient)
• Implement CopilotProvider in providers/llm/copilot.py
• Write providers/llm/prompts/copilot/deep.prompt
• Register in providers/registry.py
• Add --provider copilot to CLI args
• Write unit tests in tests/unit/providers/test_copilot_provider.py
• Update ai_slop_gate_snapshot.json (bump version)
• Update ARCHITECTURE.md provider inventory

Acceptance Criteria

• python -m ai_slop_gate.cli run --provider copilot --llm-local --policy policy.yml runs without error
• CopilotProvider returns valid ProviderObservation with observations: List[Observation]
• Cache works identically to Gemini/Groq (same CachedProvider wrapper)
• Retry logic triggers on InvalidAgentResponse
• All existing tests pass without modification

Blocked By

Copilot SDK public availability. SDK does not yet have a stable programmatic API
for non-interactive use outside VS Code. This issue should be revisited when
GitHub Copilot Extensibility exposes a
stable SDK for server-side use.

Labels

enhancement provider llm blocked