Preserve DeepSeek empty reasoning content

[ModerRAS]

Keep DeepSeek reasoning_content round-tripped even when it is an empty string, while avoiding empty reasoning patches for other providers that trigger OpenAI SDK serialization failures. Adds provider-aware gating, preserves empty reasoning in snapshot serialization, and covers the behavior with regression tests.

Summary by CodeRabbit

• New Features

  • Added configurable logging level in Config.json supporting Verbose, Debug, Information, Warning, Error, and Fatal (default: Verbose)
  • Enhanced error reporting with structured error messages across LLM and tool execution services

• Documentation

  • Updated build guide with simplified local debugging instructions
  • Added LogLevel configuration examples to README with supported values

[coderabbitai]

Caution

Review failed

Pull request was closed or merged during review

📝 Walkthrough

Walkthrough

This PR implements comprehensive structured logging improvements with configurable log levels, exception handling utilities, reasoning-content support for thinking-mode models (DeepSeek), enhanced tool-execution diagnostics across LLM and agent services, and draft ID generation refinements. Core changes include a new ExceptionExtensions class, Env-level log-level configuration, widespread application of structured error logging, and support for preserving empty reasoning content in conversation snapshots.

Changes

Logging Infrastructure & Exception Handling

Layer / File(s)	Summary
Logging Configuration & Exception Utilities Foundation TelegramSearchBot.Common/Env.cs, TelegramSearchBot.Common/ExceptionExtensions.cs	Adds Config.LogLevel property (default "Verbose"), Env.SerilogMinimumLevel static field, ResolveSerilogMinimumLevel parser (validates string to LogEventLevel); introduces ExceptionExtensions with GetPrimaryException (unwraps nested TargetInvocationException/AggregateException) and GetLogSummary (formats type name + message).
Program Logger Configuration & Tests TelegramSearchBot/Program.cs, TelegramSearchBot.Test/Common/EnvBotApiEndpointTests.cs	Updates Program.Main Serilog configuration to use Env.SerilogMinimumLevel for global and per-sink (console/file/OpenTelemetry) minimum levels; reworks OpenTelemetry sink to explicit fluent parameter setup. Adds parameterized test methods validating log-level parsing and Verbose fallback for invalid inputs.
Bootstrap Startup Error Logging TelegramSearchBot/AppBootstrap/AppBootstrap.cs	Updates TryDispatchStartupByReflection and PrintUsageHint exception handlers to use Log.Error(exception, ...) with structured context instead of embedding exception messages, providing consistent diagnostics during startup.

Reasoning Content for Thinking-Mode Models (DeepSeek)

Layer / File(s)	Summary
Reasoning Content Schema & Documentation TelegramSearchBot.Common/Model/AI/LlmContinuationSnapshot.cs	Updates SerializedChatMessage.ReasoningContent documentation to clarify that reasoning content returned by the provider should be preserved when resuming conversation state for thinking-mode models.
OpenAI Reasoning Content & Argument Sanitization TelegramSearchBot.LLM/Service/AI/LLM/OpenAIService.cs	Introduces SanitizeAndTruncateArguments helper (redacts sensitive JSON keys like api\_key/token/password, escapes newlines, truncates). Adds ShouldIncludeEmptyReasoningContent detection (returns true for DeepSeek gateway/models). Enhances GetAssistantReasoningContent with Patch-based $.reasoning_content extraction. Extends DeserializeProviderHistory to accept includeEmptyReasoningContent flag. Updates SetAssistantReasoningContent to conditionally preserve empty reasoning.
OpenAI Responses Service Tool Call & Function Normalization TelegramSearchBot.LLM/Service/AI/LLM/OpenAIResponsesService.cs	Normalizes function-call metadata (id/name/arguments JSON) with debug logging during main and resume paths. Enriches tool-execution error logging with call id, arguments, and error summary. Adds lastResolvedCallId tracking in snapshot deserialization to fill missing output call ids. Parses both __FUNC_CALL__ and __FUNC_OUTPUT__ markers with normalized metadata during resume.
Reasoning Content Test Coverage TelegramSearchBot.LLM.Test/Service/AI/LLM/OpenAIProviderHistorySerializationTests.cs	Replaces null-reasoning test with DeserializeProviderHistory_WithNullReasoningContent_Succeeds. Adds parameterized tests for ShouldIncludeEmptyReasoningContent (DeepSeek true, MiniMax false). Introduces async test for DeepSeek empty-reasoning serialization into OpenAI request. Adds CapturingHandler to validate request bodies.

Structured Logging Across LLM and Agent Services

Layer / File(s)	Summary
LLM Service Tool Execution Error Logging TelegramSearchBot.LLM/Service/AI/LLM/AnthropicService.cs, TelegramSearchBot.LLM/Service/AI/LLM/GeminiService.cs, TelegramSearchBot.LLM/Service/AI/LLM/OllamaService.cs, TelegramSearchBot.LLM/Service/AI/LLM/OpenAIService.cs	Updates AnthropicService native and XML tool-call error handling to log with tool identifiers and GetLogSummary, wrapping JSON deserialization in try/catch with empty-dict fallback. Updates GeminiService, OllamaService, and OpenAIService tool-execution catch blocks to log structured error context (tool name, arguments, error summary) and return error messages based on GetLogSummary.
LLM Controller Agent Failure Logging TelegramSearchBot/Controller/AI/LLM/GeneralLLMController.cs, TelegramSearchBot/Controller/AI/LLM/LLMIterationCallbackController.cs	Adds structured error logging to GeneralLLMController.ExecuteAsync and LLMIterationCallbackController.HandleContinue when AI agent task completes with AgentChunkType.Error, recording ChatId, MessageId/SnapshotId, and ErrorMessage.
Polling Service & Remote Tool Execution Logging TelegramSearchBot/Service/AI/LLM/ChunkPollingService.cs, TelegramSearchBot.LLMAgent/Service/ToolExecutor.cs	Adds optional ILogger<T> injection (defaulting to NullLogger) to ChunkPollingService and ToolExecutor. Wraps per-task polling in try/catch with RedisException vs general exception distinction. Enhances ToolExecutor.ExecuteRemoteToolAsync with logging at queueing, wait-error, timeout, failure, and completion phases.
Telegram Task Consumer Tool Execution Logging TelegramSearchBot/Service/AI/LLM/TelegramTaskConsumer.cs	Adds structured logging for tool execution start (request/tool/chat/user/message ids, argument keys), tool result persistence (success/length/error context), and explicit completion logs for send_message and generic tool tasks (request ID, tool name, result length).
Agent Loop Retry & Failure Diagnostics TelegramSearchBot.LLMAgent/Service/AgentLoopService.cs	Enriches transient-retry handling with computed retryReason and structured persistence of lastRetryReason/lastRetryDetails/lastRetryExceptionType. Updates terminal-failure path to compute errorSummary and persist errorType/errorDetails.
LLM Agent Program Tool Loading & Registration Logging TelegramSearchBot.LLMAgent/LLMAgentProgram.cs	Adds debug logging for Redis tool-definition loading (payload bytes). Enhances proxy tool invocation routing with informational logging and wraps remote execution in try/catch for error logging with ErrorSummary. Logs import summary with truncated tool names. Sets logger minimum level to Trace in BuildServices.

MCP Tool Execution and Registration Logging

Layer / File(s)	Summary
MCP Tool Execution with Timing and Argument Sanitization TelegramSearchBot.LLM/Service/AI/LLM/McpToolHelper.cs	Updates ExecuteRegisteredToolAsync with Stopwatch timing. Adds enriched structured logging for tool routing (chat/user/message ids, argument keys, truncated payloads with max-length constants). Implements null-safe argument trimming and CDATA marker regex stripping. Maintains enhanced pre/post logging across proxy, external, and local tool paths.
External MCP & Proxy Tool Registration with Redis Refresh Logging TelegramSearchBot.LLM/Service/AI/LLM/McpToolHelper.cs	Enhances ExecuteExternalToolAsync with metadata logging using argument formatting/truncation helpers. Adds logging when RegisterExternalMcpTools clears stale registry. Adds RefreshAgentToolDefsInRedisAsync logging with tool count and names. Updates RegisterProxyTools to track skipped-local count and emit detailed registration summary (up to 80 tool names).

Draft Message Helpers and Documentation Updates

Layer / File(s)	Summary
Draft ID Generation & Content Preview Helpers TelegramSearchBot/Service/BotAPI/SendMessageService.Streaming.cs	Replaces tick-based XOR draft-id expression with CreateDraftId helper (mixes chatId/replyTo/time/GUID with non-zero fallback). Adds GetContentPreview for normalized/truncated content summaries. Enriches SendDraftStream with structured start/debug/error logging including draft ID and content previews.
Documentation & Configuration Updates Docs/Build_and_Test_Guide.md, README.md	Updates build guide to replace command-line log-level flag examples with Config.json LogLevel default documentation. Adds LogLevel property to README.md config example with supported Serilog levels and defaults.

Estimated code review effort

🎯 4 (Complex) | ⏱️ ~75 minutes

Possibly related PRs

• ModerRAS/TelegramSearchBot#351: Adds the same ExceptionExtensions with GetLogSummary() and applies it across LLM services and agent paths for diagnostics-focused enhancements.
• ModerRAS/TelegramSearchBot#333: Both PRs advance reasoning-content handling for thinking-mode models, adding Patch-based extraction and DeserializeProviderHistory/SetAssistantReasoningContent logic for DeepSeek empty-reasoning preservation.
• ModerRAS/TelegramSearchBot#343: Both PRs modify OpenAIResponsesService tool-call and snapshot handling, with the retrieved PR wiring the service initially and the main PR enhancing its error logging and function-call normalization.

Poem

🐰 Logs now flow with crystal clarity,
Exceptions unwrapped, no mystery!
DeepSeek's thoughts preserved with care,
Structured trails of debug everywhere. ✨📋

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 34.38% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The PR title 'Preserve DeepSeek empty reasoning content' accurately reflects the main objective: enabling DeepSeek provider reasoning_content values (especially empty strings) to be preserved during serialization round-trips.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch codex/add-llm-trace-logging-and-fix-empty-encoded-value

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

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

[github-actions]

PR Check Report

Summary

• PR: Preserve DeepSeek empty reasoning content #352
• Branch: codex/add-llm-trace-logging-and-fix-empty-encoded-value <- master
• Event: pull_request
• Commit: 14265f5

Test Results

Platform	Status	Details
Ubuntu	Passed	Tests passed, artifacts uploaded
Windows	Passed	Tests passed, artifacts uploaded

Code Quality

• Code formatting check
• Security vulnerability scan
• Dependency analysis
• Code coverage collection

Test Artifacts

• Test results artifacts count: 2
• Code coverage uploaded to Codecov

Links

• PR Page
• Actions Logs

---

This report is auto-generated by GitHub Actions