v1.20.0 — pydantic-deep v0.3.3, agent.yaml, Thinking, E2E Tests

Added

• pydantic-deep v0.3.3 — upgraded from v0.2.21; thinking/reasoning, eviction, patch_tool_calls, improved context management
• agent.yaml (DeepAgentSpec) — declarative agent config via YAML in workspace; customize tools, thinking depth, checkpoints, memory in Files tab
• Thinking/reasoning — agents think before answering (thinking: low); visible in chat
• Auto-eviction — large tool outputs truncated at 5% of model context (min 5K tokens)
• context_discovery — auto-discovers AGENT.md, SKILL.md, DEEP.md, SOUL.md, CLAUDE.md
• Legacy agent migration — agent.yaml auto-created on next start
• E2E test suite — 12 Playwright tests with video/screenshots
• Guide tab updated — agent.yaml card, Thinking/Plans tools, DEEP.md/SOUL.md tips

Changed

• Model/instructions in agent.yaml always overridden by backend (free models protection)
• skill_directories format: dict → string list (pydantic-deep v0.3.x)

---

Добавлено

• pydantic-deep v0.3.3 — thinking/reasoning, eviction, patch_tool_calls
• agent.yaml — декларативная конфигурация через YAML, редактируется в Files tab
• Thinking — агент думает перед ответом
• Auto-eviction — обрезка больших output (5% контекста, min 5K)
• context_discovery — автоподхват AGENT.md, SKILL.md, DEEP.md, SOUL.md
• Миграция — agent.yaml создаётся для старых агентов автоматически
• 12 E2E тестов Playwright с видео и скриншотами
• Guide tab — карточка agent.yaml, Thinking/Plans

Изменено

• Защита от платных моделей — backend перезаписывает model/instructions