feat: enable automatic injection of the current timestamp into user prompts#3268
feat: enable automatic injection of the current timestamp into user prompts#3268no-teasy wants to merge 5 commits intoagentscope-ai:mainfrom
Conversation
…optimization ## Summary Add automatic time injection into user messages to make AI time-aware without tool calls, supporting prompt caching to reduce API costs. ## Changes ### Backend (Python) - **config.py**: Add TimeAwarenessConfig model with enabled/format fields (default disabled) - **message_processing.py**: Implement inject_time_awareness() and get_last_user_message() - Support user timezone (reuse user_timezone config) - Multi-language labels (Chinese/English auto-switch) - Custom strftime format support - Graceful fallback for invalid timezone/format (fallback to UTC) - **runner.py**: Integrate time injection into query_handler() pipeline - **config.py (routers)**: Add GET/PUT /config/time-awareness API endpoints - TimeAwarenessResponse Pydantic model for OpenAPI docs - TimeAwarenessUpdate model for request validation ### Frontend (TypeScript/React) - **timeAwareness.ts**: Add API module for time awareness CRUD operations - **TimeAwarenessCard.tsx**: Create UI component with: - Switch toggle for enable/disable - Custom format input with quick templates - Real-time preview (static example text) - Loading states and error handling - Dark mode support - **index.tsx**: Integrate TimeAwarenessCard into Agent Config page - **index.module.less**: Add styles for time awareness card - **zh.json / en.json**: Add i18n translations (25 keys each) ### Tests - **test_time_awareness.py**: Comprehensive unit tests (30+ test cases) - Configuration loading and backward compatibility - Time formatting with different timezones - Chinese/English label switching - Custom format support - Timezone fallback mechanisms - Integration with message processing - Edge cases and error handling - Performance benchmarks (< 1ms per call) ## Technical Details - Default: disabled (backward compatible, zero impact) - Injection point: Before user message content in LLM request - Format: [当前时间: YYYY-MM-DD HH:MM:SS TZ (Weekday)] or [Current time: ...] - Performance: ~160μs/call (6x faster than 1ms target) - Error handling: All paths have graceful fallbacks + logging ## Code Review Fixes Applied (P0/P1) - Fix Config loading: Use load_config() instead of load_agent_config() - Add Pydantic models: response_model + request validation for API endpoints - Format validation: Invalid strftime gracefully falls back to default - Language detection: Support zh-CN, zh-TW variants (not just 'zh') - UI preview: Show static example instead of client-side Date() ## Verification ✅ 27/27 code review checks passed ✅ 8/8 language detection test cases passed ✅ Format fallback working correctly ✅ All existing functionality preserved (regression-free) ✅ Backward compatible (old configs work without changes)
- Fix import ordering and remove unused imports - Resolve line length violations - Fix TypeScript unused import error - Apply Prettier formatting - Clean up unnecessary lambda expressions in tests - Fix config import path
refactor: 调整 load_config 函数导入路径
feat(config,agents): 提示词自动注入当前时间
github-actions
bot
added
the
first-time-contributor
Welcome to CoPaw! 🐾Hi @no-teasy, thank you for your first Pull Request! 🎉 📋 About PR TemplateTo help maintainers review your PR faster, please make sure to include:
Complete PR information helps speed up the review process. You can edit the PR description to add these details. 🙌 Join Developer CommunityThanks so much for your contribution! We'd love to invite you to join the official CoPaw developer group! You can find the Discord and DingTalk group links under the "Developer Community" section on our docs page: We truly appreciate your enthusiasm—and look forward to your future contributions! 😊 We'll review your PR soon. |
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Code Review
This pull request introduces a 'Time Awareness' feature that automatically injects the current time into user messages, enabling AI models to be time-aware without explicit tool calls while supporting prompt caching. The implementation includes a new React configuration card, backend API support, and message processing utilities with timezone and language support. Feedback suggests debouncing the format input's API calls to prevent excessive network traffic, ensuring that template selections are persisted to the server, and hardening the configuration attribute access to prevent potential attribute errors.
| const handleFormatChange = useCallback( | ||
| async (e: React.ChangeEvent<HTMLInputElement>) => { | ||
| const newFormat = e.target.value; | ||
| setFormat(newFormat); | ||
|
|
||
| if (enabled) { | ||
| setSaving(true); | ||
| try { | ||
| await api.updateTimeAwareness({ | ||
| enabled, | ||
| format: newFormat || null, | ||
| }); | ||
| message.success(t("timeAwareness.formatSaveSuccess")); | ||
| } catch (err) { | ||
| console.error("Failed to update time format:", err); | ||
| message.error(t("timeAwareness.formatSaveFailed")); | ||
| // Revert on error | ||
| setFormat(format); | ||
| } finally { | ||
| setSaving(false); | ||
| } | ||
| } | ||
| }, | ||
| [enabled, format, t, message], | ||
| ); |
There was a problem hiding this comment.
The handleFormatChange function triggers an API call on every keystroke (onChange). This can lead to a high volume of unnecessary network requests and potential race conditions if the user types quickly. It is recommended to debounce the API call or trigger it on onBlur.
| <Tag | ||
| onClick={() => setFormat("%Y-%m-%d %H:%M:%S")} | ||
| style={{ cursor: "pointer" }} | ||
| > | ||
| Default | ||
| </Tag> | ||
| <Tag | ||
| onClick={() => setFormat("%Y-%m-%d %H:%M")} | ||
| style={{ cursor: "pointer" }} | ||
| > | ||
| Compact | ||
| </Tag> | ||
| <Tag | ||
| onClick={() => setFormat("%H:%M")} | ||
| style={{ cursor: "pointer" }} | ||
| > | ||
| Time Only | ||
| </Tag> | ||
| </Space> |
There was a problem hiding this comment.
Clicking the quick template tags only updates the local format state but does not persist the change to the server. Users might expect these templates to be saved immediately, especially since the main input field attempts to save on change. Consider calling the update API within the tag click handlers.
| language = ( | ||
| getattr(config.agents, "language", "en") | ||
| if hasattr(config, "agents") | ||
| else "en" | ||
| ) |
There was a problem hiding this comment.
The code uses getattr(config.agents, "language", "en") after checking hasattr(config, "agents"). However, if config.agents exists but is None, getattr will raise an AttributeError. A safer approach is to use nested getattr calls or check if config.agents is not None.
agents = getattr(config, "agents", None)
language = getattr(agents, "language", "en") if agents else "en"
|
@no-teasy |
举个例子，AI 仅在首次调用函数时获取了当天的时间，而不会在每一轮对话中都重新获取。这就导致到了第二天，AI 依然认为还是第一天，无法识别正确的日期。所以我特意加了一个可调节的开关来解决这个问题，完全不影响正常使用 Translate in English: |
|
@no-teasy The current time is retrieved for each conversation. See https://github.com/agentscope-ai/QwenPaw/blob/main/src/qwenpaw/app/runner/utils.py#L51-L58 |
@xieyxclack However, placing the time in the system prompt seems to break the context cache (KV cache), especially with OpenAI-format models. |
|
@no-teasy So here it only includes the date, not the seconds. |
|
@no-teasy Another reason not to add it to the user prompt (apart from being redundant information) is that it will cause the console frontend to render this information. |
2 participants
Summary / 概述
Add automatic time injection into user messages to make AI time-aware without tool calls. The time string is prepended (before user content) so that the variable part (time) acts as a fixed-position prefix, keeping the user's actual message at a stable offset — this maximizes prompt cache hit rate and reduces API costs. The feature can be toggled on/off in real-time via WebUI.
为用户消息自动注入当前时间，使 AI 无需调用工具即可感知时间。时间字符串前置于用户内容之前，使可变部分（时间）作为固定位置前缀，用户实际消息保持稳定偏移量 — 从而最大化提示词缓存命中率，降低 API 成本。该功能可通过 WebUI 实时启停。
Related Issue: N/A
Security Considerations: No sensitive data involved. Timezone is read from existing
user_timezoneconfig only.Why This Design / 设计思路
Prompt Cache Optimization / 提示词缓存优化
The final request is system prompt + current time + user.
最终的请求为系统提示词+当前时间+用户提问
Important
The current timestamp is omitted from the system prompt to preserve prefix matching integrity. Since the time varies with each request, including it would cause cache invalidation and reduce cache hit rates.
系统提示词中未嵌入当前时间戳，鉴于其随请求动态变化的特性，这可以维持前缀匹配的一致性，从而确保提示词缓存的高效命中。
By placing the time string before user input (via
prepend_to_message_content), the user's message content always starts at the same position in the prompt. When users send similar messages across requests, the suffix portion is more likely to match cached entries, maximizing cache hit rate.将时间字符串通过
prepend_to_message_content前置于用户输入之前，用户消息内容始终在 prompt 中的相同位置开始。当用户跨请求发送相似消息时，后缀部分更容易命中缓存，从而最大化缓存命中率。Type of Change / 变更类型
Component(s) Affected / 影响组件
Changes / 变更详情
Backend (Python) — 4 files
src/copaw/config/config.py(+34 lines)TimeAwarenessConfig(BaseModel)withenabled: bool(defaultFalse) andformat: Optional[str]fieldstime_awarenessfield in rootConfigmodel viadefault_factory=TimeAwarenessConfigtime_awarenessload with defaultssrc/copaw/agents/utils/message_processing.py(+96 lines)inject_time_awareness(config) -> Optional[str]: Generates formatted time stringconfig.time_awareness.enabledandconfig.user_timezonezh-CN/zh-TW→ Chinese label, else English)strftimeformat with graceful fallback on invalid formatlogger.warningget_last_user_message(msgs) -> Optional[Msg]: Finds last user-role message from listNoneguard to existingprepend_to_message_content()for robustnesssrc/copaw/app/runner/runner.py(+29 lines)query_handler(), afteragent.rebuild_sys_prompt():load_config()config.time_awareness.enabled, callsinject_time_awareness()→prepend_to_message_content()try/exceptwithlogger.warning(non-critical failure)src/copaw/app/routers/config.py(+73 lines)TimeAwarenessResponse(BaseModel): Pydantic response model for OpenAPI docsTimeAwarenessUpdate(BaseModel): Request body withextra="ignore"validationGET /config/time-awareness: Returns current settingsPUT /config/time-awareness: Updatesenabled/formatand persists viasave_config()Frontend (TypeScript/React) — 6 files
console/src/api/modules/timeAwareness.ts(new, +17 lines)TimeAwarenessConfiginterface +getTimeAwareness()/updateTimeAwareness()API callsconsole/src/api/index.ts(+4 lines)timeAwarenessApiin unifiedapiobjectconsole/src/pages/Agent/Config/components/TimeAwarenessCard.tsx(new, +220 lines)strftimeinput withallowClear, validated server-sideDate()inconsistency)get_current_time()tool when offindex.module.lessconsole/src/pages/Agent/Config/components/index.ts(+1 line)console/src/pages/Agent/Config/index.tsx(+3 lines)TimeAwarenessCardin Agent Config pageconsole/src/pages/Agent/Config/index.module.less(+115 lines):global(.dark-mode)override supportconsole/src/locales/zh.json/en.json(+25 keys each)Tests — 1 file
tests/test_time_awareness.py(new, +368 lines, 30+ test cases)TestTimeAwarenessConfig: Default values, custom format, JSON serializationTestConfigIntegration: Field presence in root Config, backward compatibility, enable from JSONTestInjectTimeAwareness: Chinese/English labels, custom format, timezone fallback, disabled state, error handlingTestGetLastUserMessage: Normal case, empty list, no user message, mixed rolesTestPrependToMessageContent: String content, list content,Noneguard (new)Technical Details / 技术细节
enabled: False— zero impact on existing deployments / 默认关闭，对现有部署零影响prepend_to_message_content()— before user content / 通过prepend_to_message_content()前置于用户内容之前config.user_timezone(IANA format) / 复用现有config.user_timezone（IANA 格式）config.agents.language:zh*→ Chinese label, else English / 自动检测config.agents.languagestrftimestring, falls back to%Y-%m-%d %H:%M:%S {tz} ({weekday})/ 可选strftime字符串，降级为默认格式GET/PUT /config/time-awarenesswith Pydantic models for validation / Pydantic 模型校验Checklist / 检查清单
pre-commit run --all-fileslocally and it passespytestor as relevant) and they passTesting / 测试方式
curl -X PUT http://localhost:8000/config/time-awareness -H 'Content-Type: application/json' -d '{"enabled": true}'%H:%M→ verify compact outputuser_timezonevalues (e.g.,America/New_York,Europe/London)config.jsonwithouttime_awarenessloads normallyLocal Verification Evidence / 本地验证结果
Additional Notes / 补充说明
TimeAwarenessConfigdefaults toenabled: False, existing configs work without changesTimeAwarenessConfig默认enabled: False，现有配置无需修改config.jsonviasave_config(), survives restartssave_config()持久化到config.json，重启后保留设置