Summary

• Adds GenerationConfigOverrides, ReasoningConfigOverride, and LlmOverrides models to FineGrainedConfigHeader
• Enables per-LLM generation config overrides (temperature, reasoning, max_completion_tokens, etc.) via the llm section of the X-Config header
• Mirrors secure-orchestrator PR #392

Usage

from sequrity.control import FineGrainedConfigHeader, LlmOverrides, GenerationConfigOverrides, ReasoningConfigOverride

config = FineGrainedConfigHeader(
    llm=LlmOverrides(
        pllm=GenerationConfigOverrides(
            temperature=0.5,
            reasoning=ReasoningConfigOverride(enabled=True, effort="high"),
        ),
        rllm=GenerationConfigOverrides(max_completion_tokens=512),
    ),
)

Test plan

• 7 new unit tests for serialization, reasoning config, multiple LLMs, None exclusion, deep merge, and strict validation
• All 34 header override tests passing

🤖 Generated with Claude Code