Fix/gemini thinking mode

[sartorijr92]

This pull request addresses several issues and improvements in the kodustech/kodus-ai repository, specifically targeting the Gemini thinking mode. Key changes include:

1. Error Handling and Optional Chaining: Introduces optional chaining in generate-issues-from-pr-closed.use-case.ts to prevent null reference errors when accessing organizationId. However, it highlights a potential issue where a cache-clearing function might be called with undefined, suggesting the need for a conditional check.

2. Custom String Output Parser: Refactors multiple services to replace the standard StringOutputParser with a CustomStringOutputParser. This change is applied across various files, including commentManager.service.ts, crossFileAnalysis.service.ts, llmAnalysis.service.ts, promptRunner.service.ts, kodyIssuesAnalysis.service.ts, kodyRulesAnalysis.service.ts, kodyRulesPrLevelAnalysis.service.ts, and codeASTAnalysis.service.ts. The custom parser aims to enhance compatibility, robustness, and specialized handling of LLM-generated JSON output.

3. LLM Provider Enhancements: Updates in llmModelProvider.helper.ts and llmProvider.service.ts include replacing ChatGoogleGenerativeAI with ChatGoogle for Gemini models and introducing a maxReasoningTokens parameter. This parameter is integrated into model configurations and error logging metadata, with a fallback to default values.

4. Logical Flaw in Custom Parser: The newly introduced CustomStringOutputParser in langchainCommon/customStringOutputParser.ts contains a logical flaw, where it incorrectly accesses a property on an array, affecting the filtering of 'reasoning' content blocks.

Overall, this pull request focuses on improving error handling, enhancing parsing logic, and updating LLM provider configurations to support the Gemini thinking mode.

[kodus-ai]

Changes were made to the Google Gemini integration in llmModelProvider.helper.ts, which uses the process.env.API_GOOGLE_AI_API_KEY variable. Please check with @wellington to confirm if any updates are required for this environment variable in the self-hosted environment due to these changes. Kody Rule violation: Verificar atualizações de variáveis de ambiente

[kodus-ai]

maxReasoningTokens: options.maxReasoningTokens,

The code attempts to access strategy.maxReasoningTokens within a block where strategy is known to be undefined. The if (!strategy) condition on line 55 ensures that this code path is only executed when strategy is falsy. Accessing a property on an undefined object will cause a TypeError, crashing the error logging process. The logger should only reference options.maxReasoningTokens since the strategy-based fallback is not available in this context.

_{Talk to Kody by mentioning @kody}

_{Was this suggestion helpful? React with 👍 or 👎 to help Kody learn from this interaction.}

​

​

[kodus-ai]

        const filteredContent = content.filter((item: { type: string }) => item.type !== 'reasoning');
        return super._messageContentComplexToString(filteredContent);

The content parameter is of type MessageContentComplex, which is an array of content blocks (e.g., {type: 'text', ...}). The current code content?.type attempts to access a type property on this array, which will always be undefined. As a result, the condition content?.type === 'reasoning' is never met, and the custom logic has no effect.

To correctly handle 'reasoning' content blocks, you should filter the content array to remove items of that type before passing the array to the parent method for string conversion. This will correctly remove reasoning steps while preserving other useful content.

_{Talk to Kody by mentioning @kody}

_{Was this suggestion helpful? React with 👍 or 👎 to help Kody learn from this interaction.}

​

​

[kodus-ai]

Found critical issues please review the requested changes

• Relying on a custom parser instead of the LLM provider's native JSON mode makes JSON parsing less reliable.
• Calling clearIssuesCache with a potentially undefined value may lead to unintended side effects and compromises data traceability.
• A TypeError will occur when logging an error for an unsupported provider because strategy is accessed while it is undefined.
• The logic incorrectly checks for a type property on an array, causing the intended filtering of 'reasoning' content to never occur.

[kodus-ai]

Kody Review Complete

Great news! 🎉
No issues were found that match your current review configurations.

Keep up the excellent work! 🚀

Kody Guide: Usage and Configuration

Interacting with Kody

• Request a Review: Ask Kody to review your PR manually by adding a comment with the @kody start-review command at the root of your PR.

• Provide Feedback: Help Kody learn and improve by reacting to its comments with a 👍 for helpful suggestions or a 👎 if improvements are needed.

Current Kody Configuration

Review Options

The following review options are enabled or disabled:

Options	Enabled
Security	✅
Code Style	❌
Kody Rules	✅
Refactoring	✅
Error Handling	❌
Maintainability	✅
Potential Issues	✅
Documentation And Comments	❌
Performance And Optimization	✅
Breaking Changes	✅

Access your configuration settings here.

​

[kodus-ai]

The introduction of the maxReasoningTokens parameter in llmProvider.service.ts suggests a new configuration may be required. Per our process for environment-related changes, please mention @wellington to verify if this new setting needs to be configured in the self-hosted environment's variables (e.g., in the .env file). Kody Rule violation: Verificar atualizações de variáveis de ambiente