cockpit v2.0: agent loop with tool calling (LLM decides → execute → loop)

Replace hardcoded intent routing with agent loop:
- 14 tool definitions injected into prompt
- LLM decides which tools to call and in what order
- Multi-turn: one message can trigger switch_project + create_task + add_repo + agent_status
- Max 8 turns, security scan kept, question routing preserved
- 1200→600 lines, all business logic in tool executors
This commit is contained in:
ymq 2026-08-08 15:40:34 +08:00
parent 4c5b3104cf
commit 282474898c

File diff suppressed because it is too large Load Diff