AI/Tasks/CurrentTask.txt
|
Your task is to implement full support for OpenAI GPT-5.3-Codex reasoning
effort modes in the TechToolbox.Agent codebase. Target directory: C:\repos\TechToolbox\src\TechToolbox.Agent Reasoning effort modes to support: low medium high xhigh These correspond to the "reasoning.effort" field in the /v1/responses API. You MUST perform the following actions: 1. Modify AgentConfiguration.cs to add: - ReasoningEffortOverride (string? or enum) - EnableReasoningEffortAuto (bool) - EffectiveReasoningEffort (string? or enum) 2. Modify the PowerShell/CLI parameter ingestion layer so the user may specify: - -ReasoningEffort <low|medium|high|xhigh> - -ReasoningEffortAuto (switch) 3. Implement a new policy method in AgentOrchestrator.cs: ResolveReasoningEffort(AgentConfiguration config) This method must: - Respect explicit override - Use auto mode when override is absent - Map preflight score to effort level - Optionally incorporate ThinkingMode - Store the final value in config.EffectiveReasoningEffort 4. Modify OpenAiCompatibleLlmClient.cs so that the Responses API payload includes: reasoning: { effort: "<value>" } ONLY when provider=openai AND model=gpt-5.3-codex AND effort is non-null. 5. Ensure the reasoning effort value is threaded through the entire call chain: - AgentOrchestrator → LLM client → BuildPayload → SendRequestAsync 6. Add diagnostics: - Selected reasoning effort - Whether override or auto was used - Whether the field was emitted to the payload 7. Add guardrails: - If provider/model does not support reasoning effort, omit the field and log a diagnostic. 8. For EVERY file you modify, you MUST output a WRITE-FILE block with: - The full absolute path - The complete updated file contents - No placeholders - No partial patches - No diff format - The entire file rewritten exactly as it should exist on disk Example format: WRITE-FILE Path: C:\repos\TechToolbox\src\TechToolbox.Agent\Configuration\AgentConfiguration.cs <full updated file contents here> 9. After all WRITE-FILE blocks, output a final markdown summary describing: - All files modified - All new fields added - The reasoning-effort policy - The payload changes - Any new parameters - Any new diagnostics - Confirmation that GPT-5.3-Codex reasoning effort is now implemented. This task is complete only when: - All modified files have been emitted via WRITE-FILE blocks - The codebase is fully updated - The markdown summary reflects the completed implementation |