feat: Microsoft Agent Framework compatibility and bypass mode

[CaddyGlow]

Summary

• Added OpenAI Responses format support to mock handler with proper SSE event sequences
• Added format-aware mock adapter with target format detection from route context
• Added bypass mode to plugin factory with MockAdapter fallback for testing
• Added inject_detection_payload config toggle for generic OpenAI-compatible API usage
• Added openai_thinking_xml streaming configuration support

PR Stack

This is PR 2 of 3 split from #41. Stacked on #43. Merge in order:

1. feat: Codex v0.114.0 integration with gpt-5.4 support #43 - Core Codex v0.114.0 integration
2. This PR - MSAF compatibility and bypass mode
3. feat/codex-websocket-hardening - WebSocket auth hardening and e2e tests

Test plan

• All pre-commit checks pass (ruff, mypy, format)
• 1101 unit/plugin tests pass (+20 over PR1)
• MSAF integration and real library tests pass
• Mock handler Responses format streaming emits valid OpenAI events
• Bypass mode factory tests verify adapter creation and error handling

[Copilot]

Pull request overview

This PR extends ccproxy’s Codex/OpenAI compatibility layer to better support Microsoft Agent Framework (MSAF) style traffic by making bypass/mock behavior format-aware (OpenAI Chat vs OpenAI Responses vs Anthropic), adding a bypass-mode adapter fallback at plugin factory level, and propagating openai_thinking_xml streaming behavior through the formatter pipeline.

Changes:

• Add OpenAI Responses-format support to the mock handler (including SSE event sequencing) and make the mock adapter resolve target format from request.state.context.format_chain / route endpoint.
• Add plugin factory bypass-mode support that returns a MockAdapter (with clear logging/error behavior).
• Add inject_detection_payload toggle to Codex config/adapter to prevent Codex CLI detection/template injection for generic OpenAI-compatible clients; propagate openai_thinking_xml via formatter context.

Reviewed changes

Copilot reviewed 19 out of 20 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
uv.lock	Dependency lock updates (notably mcp bump and additional wheels).
ccproxy/services/mocking/mock_handler.py	Adds OpenAI Responses mock conversion + Responses SSE streaming event sequence; prompt text extraction.
ccproxy/services/adapters/mock_adapter.py	Format-aware mock routing (prefers format_chain, falls back to endpoint detection); passes prompt text through.
ccproxy/services/adapters/format_adapter.py	Adds streaming configuration and propagates openai_thinking_xml via formatter contextvar.
ccproxy/services/factories.py	Wires openai_thinking_xml config into core adapters and mock handler adapters (chat + responses).
ccproxy/plugins/codex/config.py	Introduces inject_detection_payload Codex setting.
ccproxy/plugins/codex/adapter.py	Skips detection/template injection when configured; preserves user reasoning; removes additional unsupported keys.
ccproxy/llms/formatters/context.py	Adds ContextVar helpers for openai_thinking_xml propagation.
ccproxy/llms/formatters/openai_to_openai/streams.py	Respects openai_thinking_xml when emitting wrapped thinking blocks in streaming conversions.
ccproxy/llms/formatters/openai_to_openai/responses.py	Respects openai_thinking_xml when wrapping thinking blocks in non-streaming conversions.
ccproxy/core/plugins/factories.py	Adds bypass-mode factory path that returns MockAdapter + warning log.
.ccproxy.codex.msaf.toml.example	Example config for MSAF clients (disables detection payload injection, disables thinking XML).
tests/unit/services/test_mock_adapter.py	Verifies target format resolution via format_chain and endpoint fallback.
tests/unit/services/mocking/test_mock_handler.py	Adds tests for prompt extraction and validates Responses SSE event schema/order.
tests/unit/llms/test_llms_streaming_settings.py	Tests openai_thinking_xml propagation + ContextVar task isolation.
tests/unit/core/test_provider_factory_bypass.py	Tests bypass-mode factory behavior (warning + clear error without service container).
tests/plugins/codex/unit/test_adapter.py	Tests inject_detection_payload=False behavior and reasoning preservation.
tests/plugins/codex/integration/test_msaf_real_library.py	Integration coverage for sequential agent-style MSAF flows without CLI injection.
tests/plugins/codex/integration/test_msaf_compat.py	Integration coverage for MSAF chat completions compatibility + thinking XML disabled behavior.
tests/plugins/codex/integration/test_codex_basic.py	Adds bypass streaming test validating OpenAI Responses SSE events.

Comments suppressed due to low confidence (1)

ccproxy/services/adapters/mock_adapter.py:116

• RequestContext is created with a fixed request_id="mock-request". Now that mock streaming for OpenAI Chat/Responses derives event IDs (chatcmpl-*, resp_*, msg_*) and X-Request-ID from ctx.request_id, this makes IDs collide across requests in bypass/mock mode and can break clients that expect uniqueness. Consider deriving the request id from an incoming header (e.g., X-Request-ID) when present, otherwise generate a unique value (uuid4), and use it consistently for both handle_request and handle_streaming.

        ctx = RequestContext(
            request_id="mock-request",
            start_time=time.perf_counter(),
            logger=structlog.get_logger(__name__),
        )

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.