Spaces:
Sleeping
Sleeping
Commit History
-
Fix actor display and JSON snapshot parsing
yanggan- memory_helper: render source side symmetrically with target so SYSTEM/ALL get translated to 系统/全体 instead of leaking raw IDs in event lines - map_provider: add missing json_repair import and use it for snapshot parsing, so malformed JSON is repaired instead of silently failing with NameError - minor wording tweaks in prompts/tools docstringsf5eb3c9
-
Update event summary formatting in memory_helper.py
ZeroSheepfa446bf
-
Rename misleading fallback comment and simplify event summary format
ZeroSheep- Rename "fallback" comment in evaluate_trade.py to accurately describe it as a normal flow path for free trade new proposals - Add phase_round_start to ALWAYS_KEEP_STATUSES so round start events are no longer collapsed in summaries - Simplify collapsed event prefix to "[n条消息被折叠]" with detailed label breakdown preserved26a9aac -
Clean up dead code and eliminate duplication
ZeroSheepDead code removal: - Remove unused imports (json, os, AgentBuilder, ChinatownMemory, AgentReq/AgentResp) - Remove unused logging/logger in 4 handler and registry files - Delete dead build_prompt(), MapProvider.render() - Delete write-only _snapshot_version variable - Remove unreachable stale_hint branch in map_sync Duplication elimination: - Extract _update_snapshot() and _record_history() in chinatown_agent.py - Extract build_trade_fields() for trade asset serialization - Extract render_single_event() replacing 4 duplicated dispatch call sites - Extract DEAL_VERBS constant for deal verb mapping - Move shop_legend persistence to query_board() unified entry point Bug fix: - Fix _tool_calls_to_resp type annotation (AgentResp → AgentReq)c525e96 -
Refactor accept_deal to accept dealIds list in single call
ZeroSheepInstead of requiring LLM to call accept_deal multiple times in parallel for batch acceptance, the tool now takes a dealIds list parameter directly. Removed PARALLEL_OK machinery from tool_loop accordingly.f50cb01
-
Update initiate_trade and map_provider
yanggancd9979b -
Update game rules, trade tools, map provider and skip tool
yanggan82779bb
-
Update evaluate_trade handler
yanggan6a247bc -
Update handlers and tools for trade evaluation
yanggan4c4799d -
更新各 handler 和 tool_loop
yanggan0c26117 -
重构 map_sync 工具模块，更新各 handler 和 tools
yangganeaa6bf3 -
优化建筑规则提示，修复 map_provider，增强 prompt_builder 和 tool_loop
yanggan45325ca -
更新 tool_loop 多模态支持，优化交易相关工具和游戏规则提示
yangganb43c0f9 -
设置 MODEL_NAME 为 gemini-3-flash-preview
yanggan4cbeb99
-
Update tools helpers, skip and map provider
yanggan1ffee80 -
Update tools init
yanggan63911ee -
Update agent, handlers, prompts, tools and asset manager
yanggan741d850 -
refactor(map_provider): 统一棋盘快照解析接口并移除旧视觉解析模块
yanggan9577695 -
refactor(tools): 重构工具模块架构以简化维护和扩展实现
yangganca511ce
-
Update app and agent code
yanggan41b98e4 -
Add tool-calling support and update handlers
yanggan30b79a5 -
refactor(agent): 引入 parse_cash_str 函数以支持现金字符串解析
yanggan- 从 chinatown_memory 中导入 parse_cash_str 函数 - 保持原有 ChinatownMemory 导入不变 - 为后续代码现金字符串处理做准备d44e00f
-
chore: bump werewolf-agent-build-sdk to >=0.0.16 (AgentAction rename)
yanggane39b988 -
Refactor handlers, memory helper, board renderer and asset manager
yanggana72c068
-
Refactor handlers and extract vision parsing
yanggan731e66c -
Remove Gradio UI and simplify app
yanggand0cb3a9
-
Update README, app and Gradio UI
yanggan7729f38 -
Add vision model support and Gradio UI panel
yanggan- Board renderer: dual-route snapshot parsing (JSON + base64 image via vision API) - Agent: track current decision state and snapshot image for UI - App: launch Gradio visualization panel in background thread - Add gradio and pillow dependencies5b4afa4
-
refactor(select_plots): 移除冗余坐标信息处理优化代码
yanggan68726ba
-
refactor(board_renderer): 简化棋盘渲染逻辑并支持动态尺寸
yangganbdd5817 -
feat: add req/resp history endpoint, fix player_id dynamic binding, remove coordinate info from select_plots
yanggan06700a4 -
refactor(select_plots): 优化选择地块的决策处理逻辑
yanggand977280
-
Set player_id from req.target on PHASE_START
yanggan71adc85 -
Update SDK to 0.0.15
yangganeaa8f13 -
Update SDK to 0.0.14 (AgentReq now includes snapshot field)
yanggana593821 -
Pin dependency versions to resolve pip backtracking
yanggan7ab66d2 -
Update SDK to 0.0.13 (removed langchain dependency)
yangganb1b18f7 -
Initial commit: Chinatown game agent
yangganDocker-based HuggingFace Space with LLM agent for Chinatown board game.bb5c778 -
initial commit
9a1c106