M1: System Prompt Passthrough for OpenAI-Compatible API

[kovtcharov]

Priority: P0 — Blocks all webapp integrations

Effort: Small (1-2 days)

Problem

The OpenAI API's /v1/chat/completions endpoint ignores or overrides system role messages from the client. Each agent class generates its own system prompt via _get_system_prompt(), and the client has no way to inject custom context (e.g., product documentation, company knowledge base, domain-specific instructions).

Deliverable

When a client sends a system message in the messages array, GAIA should honor it. Two modes:

• Append (default): Client's system message is appended to the agent's built-in system prompt. The agent retains its core behavior (tool use, reasoning patterns) while gaining the client's additional context.
• Override: If a request header or parameter signals override mode, the client's system message replaces the agent's default entirely. For advanced integrations that want full control.

Files to change

• src/gaia/api/openai_server.py — detect and forward system messages
• src/gaia/agents/base/agent.py — _build_messages() to merge or replace system prompt

Acceptance criteria

• Client can send {"role": "system", "content": "You are a helpful assistant for XYZ..."} and the agent's responses reflect that context
• Agent's core tool-calling behavior is preserved in append mode
• Existing behavior unchanged when no system message is provided

Context

Full milestone plan: docs/plans/webapp-integration.md