Add Pydantic validation model for config.json

[Copilot]

config.json was loaded as a raw dict and accessed via .get() chains throughout the codebase, deferring misconfiguration errors to runtime callsites rather than surfacing them at startup.

New: hydra/config_schema.py

Typed Pydantic v2 models for every config section:

• FrameworkConfig — log_level validated against Python's logging set
• DashboardConfig — port bounded ge=1, le=65535
• ContextBusConfig, DatabaseConfig, ProfileConfig
• Discriminated-union adapter configs keyed on "type" (MockAdapterConfig, DiscordAdapterConfig, OpenAIAdapterConfig, MCPAdapterConfig) — with field-level constraints (e.g. temperature capped to the OpenAI API range 0.0–2.0)
• HydraConfig — top-level composite with safe defaults for every section

# Bad config now fails immediately at startup with a clear ValidationError
HydraConfig.model_validate({"framework": {"log_level": "VERBOSE"}})
# pydantic_core.ValidationError: log_level must be one of {'DEBUG', 'INFO', ...}

hydra/main.py

• load_config() return type changed from dict to HydraConfig; parses via model_validate and catches ValidationError specifically
• run() and main() use typed attribute access (config.dashboard.enabled, config.framework.log_level, etc.)

hydra/core/runtime.py

• HydraRuntime.__init__ signature updated to HydraConfig
• All .get() chains replaced with direct attribute access
• Adapter configs passed to existing constructors via model_dump() to preserve the adapter interface

[Copilot]

Pull request overview

This PR introduces a typed, validated configuration layer (Pydantic v2) for hydra/config.json and updates the runtime/entrypoint to use the validated model instead of raw dict access, so configuration errors surface at startup.

Changes:

• Added hydra/config_schema.py with Pydantic models for config sections and adapter discriminated unions.
• Updated hydra/main.py to load/validate config into a HydraConfig and use typed attribute access.
• Updated hydra/core/runtime.py to accept HydraConfig and replace .get() chains with typed access (dumping adapter configs back to dict for existing adapter constructors).

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 3 comments.

File	Description
hydra/config_schema.py	Adds Pydantic schema models for full config validation and typed access.
hydra/main.py	Parses config.json into HydraConfig and uses typed config fields.
hydra/core/runtime.py	Updates runtime to consume HydraConfig and uses model-backed values throughout initialization.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

FrameworkConfig.validate_log_level() hard-codes a limited set of level names. This will reject valid logging level aliases like WARN/FATAL (and NOTSET), which were previously accepted by setup_logging() and are part of Python’s logging level name mapping. Consider validating against logging._nameToLevel (or a curated list derived from it) and make the error message deterministic (avoid embedding a set, whose ordering is not stable).

[Copilot]

The new schema models don’t forbid unknown/extra keys, so typos in config.json (e.g., dashbaord) will be silently ignored by Pydantic’s default extra handling, which undermines the goal of surfacing misconfiguration early. Consider setting model_config = ConfigDict(extra="forbid") on HydraConfig (and/or the section models) so unexpected keys produce a ValidationError.

[Copilot]

load_config() now catches ValidationError, but invalid JSON (e.g., a trailing comma) will still raise json.JSONDecodeError and crash with a stack trace rather than a clear startup error. Consider catching json.JSONDecodeError (and potentially OSError) to log a concise message including the config path before exiting.