perf(llm): pass pre-loaded cfg dict to LLM clients, skip redundant file reads

[CGFixIT]

Problem

gate.py loads and parses config.yaml once at startup into cfg. It then instantiates LocalLLMClient() and GrokClient(), each of which independently opens and parses config.yaml from disk again — 2 extra file reads that produce the same data already in memory.

Fix

Add an optional cfg: Optional[dict] parameter to both client constructors:

# Before
class LocalLLMClient:
    def __init__(self, config_path: str = "config.yaml"):
        with open(config_path, encoding="utf-8") as f:
            cfg = yaml.safe_load(f)

# After
class LocalLLMClient:
    def __init__(self, config_path: str = "config.yaml", cfg: Optional[dict] = None):
        if cfg is None:
            with open(config_path, encoding="utf-8") as f:
                cfg = yaml.safe_load(f)

gate.py passes the already-loaded dict at both call sites:

local_llm = LocalLLMClient(cfg=cfg)
grok = GrokClient(cfg=cfg)

Backward Compatibility

Fully backward compatible — config_path fallback is unchanged. Any existing code (tests, standalone scripts) that calls LocalLLMClient() or GrokClient() without a cfg argument continues to work exactly as before.

Files Changed

• llm/client.py — optional cfg param on both constructors
• gate.py — pass cfg=cfg at the two instantiation sites

---

Generated by Claude Code