Skill 详情
travel-agents
Focused on simulated agent journeys and postcard generation rather than practical trip planning.
使用前先检查
自动化审核只检查相关性,不代表安全审查或推荐。使用前请阅读来源中的说明。
SKILL.md
这段内容是审核时保存的快照。外部来源才是完整且最新的版本。
--- name: travel-agents description: "Plan and run a travel agents journey, and directly generate strict postcard-style travel images when the user asks for postcards/images for specified places. Supports origin/destination setup, recurring Agent traveler identity, route scenes, prompts, generated postcard images, and route data export." --- # travel agents Use this skill when the user wants an AI Agent to "travel for me" and maintain a multi-day journey along a route. This is a portable coding-agent skill for UI and CLI hosts. Work through local files and scripts. ## Capability Adapters Different coding agents expose different capabilities: native image generation, weather/search, reminders/automation, browser tools, or local command execution. Select the available adapter at run time and keep trip state portable across hosts. Supported adapter categories: - Native image tool adapter: call the host agent's image-generation tool directly from the current turn when it is available, then update local state if the tool returns a local image path. - API provider adapter: local scripts generate through configured API keys or command hooks. - Automation adapter: create the daily run automation by default after a durable trip is initialized. Use host agent reminder/automation first when images require the host agent's native image tool. Use OS schedulers only when local scripts can complete image generation without the host agent through a configured local API provider or `TRAVEL_AGENTS_IMAGE_COMMAND`. - Weather adapter: host weather/search tool first when available; otherwise use a user-provided weather summary or a configured weather command. `detect_environment.py` cannot see host-platform automation tools. The coding agent must inspect its currently available tools before falling back to OS schedulers. In Codex, use the `automation_update` tool to create a real platform automation; do not simulate it with shell commands, cron text, or hand-written automation files. ## Image Capability Gate This skill's core output is generated postcards. Before first-run setup, route planning, trip initialization, or daily prompt production, verify that at least one real image-generation path is available: - A host-native image-generation tool exposed in the current agent turn. - A configured local API provider or command hook reported by `python scripts/detect_environment.py --json` under `provider.selected_api_provider`. If neither path is available, stop immediately and ask the user to provide one of them. Do not ask for origin, destination, Agent identity, day count, or start producing route data/prompts. This is blocking, not a degraded prompt-only mode. Use this concise blocking response shape: ```text 我现在还不能开始这段旅程:当前 agent 没有可用的图像生成工具,本地也没有检测到可用的图片 API provider。 请先提供其中一种能力: 1. 换到带图像生成工具的 agent/环境; 2. 配置一个本地图片 provider(例如 OPENAI_API_KEY、GEMINI_API_KEY、SEEDREAM_API_KEY 或 TRAVEL_AGENTS_IMAGE_COMMAND); 3. 直接给我一个可调用的图片生成命令。 准备好后我再开始问出发地、目的地和旅行者形象。 ``` ## Overview `travel agents` turns a user's travel wish into a local, multi-day Agent journey. The agent acts as a small recurring traveler moving from an origin to a destination. For each day, the workflow chooses a waypoint, builds a scene prompt with recognizable local details, can optionally generate an image, and advances the trip state. The route is a visually coherent, geographically plausible narrative arc for daily postcard-style scenes. The main deliverable is durable journey state, route data, daily scene prompts, and generated postcard images. ## Direct Postcard Image Requests Use this fast path when the user explicitly asks to generate one or more postcards/images from given places, an existing route, or a supplied character reference, and does not ask to start or advance a durable trip. This fast path is complete by itself when a host-native image-generation tool is available. Do not read `references/prompt_contract.md`, `references/route_planning.md`, `references在 GitHub 阅读完整来源 (打开外部页面)