功能描述

轻量版区域级拍星推荐技能，适用于今晚或指定日期的全国、省市级拍星条件筛选。默认全国扫描，先按云量粗筛，再结合真实天气细筛，输出逐晚推荐和完整气象信息。

使用说明 (SKILL.md)

GoStargazing

Name: Go Stargazing
Author: clhwbd

定位

轻量版、区域级拍星推荐 skill，用于回答：

今晚或指定日期，全国或指定范围哪里更适合拍星
多天查询时，每一晚分别哪里更适合拍星

不做：

具体机位推荐
编造真实视宁度、实测光污染、真实遮挡
把区域级结论伪装成机位级结论

默认规则

用户说“全国 / 中国 / 国内”时，默认按中国 34 个省级行政区全覆盖处理
用户指定省份或更小范围时，按用户范围查询
用户未说明题材时，默认按星空 / 银河处理

核心流程

1. 全国粗筛

大范围查询时，第一轮只看云量
云量不达标直接过滤，达标后再进入细筛
粗筛阶段不混入风速、湿度、月光、露点、降水、天气代码等指标
全国粗筛统一只用 ecmwf_ifs 看云量
单日 coarse survivor \x3C= 10：直接全部进入细筛
多天 coarse survivor \x3C= 20：直接全部进入细筛

2. 细筛

细筛阶段默认启用：

风速
湿度
夜间通透度 / 能见度
月光影响
月升 / 月落
露点
降水
天气现象码

模型策略：

粗筛与细筛统一只用 ecmwf_ifs
不再做联合复核流程

评分原则：

云量权重最高
风速权重较低

3. 区域聚合

将高分点聚合成区域级结论
优先输出自然地理或行政区域名称
不直接返回大量采样点给用户

4. 候选榜单规则

按最终排序分从高到低排序
候选阈值：max(70, 第一名分数 - 10)
只允许 recommended / backup 进入候选榜单
observe_only 不得进入候选榜单
达标候选少于 10 个时如实返回
达标候选多于 10 个时只保留前 10
不允许为了凑满 10 个而硬塞明显低质量候选

多天规则

多天查询时：

按逐晚独立推荐输出
每晚给出主推荐、备选和完整气象指标
不做跨晚路线连续性判断
只做区域级逐晚判断，不扩展为路线规划

输出要求

默认按以下顺序组织：

结论
备选
细化说明
风险提示
本轮参考信息
下一步（是否继续细筛）

真实数据输出

当用户明确要求真实数据或要自行复核时，回复中必须给出本轮实际查到的关键气象信息，不得只给口语化结论。

优先包含：

天文夜窗、最佳可拍窗口、可拍时长
平均 / 最差云量
风速、阵风、湿度、能见度
温度、露点、月光影响
低 / 中 / 高云、降水、天气现象码（如有）
数据来源与数据完整性说明

回复生成约束

调用方应优先使用 final_reply_draft / reply_drafts.* 作为最终回复，不要自行根据零散 JSON 字段二次摘要或重组，否则容易丢失完整气象指标与解释。

用户明确要求复核或对照 Windy 时，应优先使用 final_reply_draft 或 reply_drafts.detailed。

导出边界

本 skill 仅返回文本或 JSON 结果
不内置 Excel / xlsx 导出能力
如需表格附件，应由调用侧另行生成

用户可读性规则

只说自然中文
不直接输出内部字段名、调试字段或变量名
解释时优先说明：哪个区域更稳、哪个区域更适合守候、哪晚更适合当路线锚点

夜窗与月光

地区天文夜窗和最佳可拍窗口必须分开表达
月光建议可以说强 / 中 / 弱，但不要伪装成真实天文实测
月升 / 月落与无月光窗口属于辅助信息，不是硬评分主字段

运行前提

依赖 python3
额外 pip 依赖：lunarcalendar、shapely
真实天气模式依赖可访问的 Open-Meteo 在线接口
若当前环境无法访问外部天气源，或当天请求额度已耗尽，则必须明确告知用户无法继续获取新的真实天气数据

数据边界

天气主源：Open-Meteo HTTP API
需要真实数据时，必须使用真实天气查询（CLI 对应 --real-weather），不得退回 mock 后继续按真实结果回复用户
若真实天气抓取失败，必须明确说明失败情况，不得用 mock 顶替真实结论
预报范围最远约 16 天，超出需明确告知用户无法查询
复杂地形区域（高原 / 山区）预报不确定性更高，必要时提醒用户临近出发再复查
命中 429 时应自动退避，不得继续高频请求同一模型
若天气源明确返回 Daily API request limit exceeded. Please try again tomorrow.，则视为当天真实天气额度已耗尽
命中上述“日额度超限”时：不得继续重试；应直接对用户说明：今天天气源查询额度已超限，明天再查
不同天气模型要分别做节流，不能共用一条激进并发节奏
全国扫描与单区复核必须分开：全国允许较高吞吐，小范围 / 单区复核应改走低频保守模式

打包数据

当前运行使用：

data/china-provinces-lite.json
data/china-prefectures-lite.json

使用轻量 bbox 边界，替代完整 geometry，以控制包体积。

主要脚本

scripts/go_stargazing.py：主入口
scripts/go_stargazing_engine/engine.py：主流程编排
scripts/go_stargazing_engine/geo.py：地理边界与空间处理
scripts/go_stargazing_engine/weather.py：天气获取与聚合
scripts/go_stargazing_engine/scoring.py：评分逻辑
scripts/go_stargazing_engine/astronomy.py：天文计算

致谢

本技能由以下伙伴共同打造：

陈一菲：Skill 第一版创作者，为星空摄影规划奠定了核心框架
陆不喝：提供技术方案与专业指导，让数据驱动的选址评分体系成为可能
Fortune：发起者，将摄影实践与 AI 能力深度结合，持续推动功能迭代

感谢每一位愿意为星空奔赴山海的人

安全使用建议

This skill appears coherent and implements the behavior it describes. Before installing: (1) be aware it requires Python3 and pip-installing lunarcalendar and shapely (shapely can need system packages on some platforms); (2) the skill will make outbound HTTP requests to Open‑Meteo when you request real-weather mode — ensure you are comfortable allowing outbound network access and that you respect Open‑Meteo's usage limits and terms; (3) national (wide) scans can trigger many concurrent requests and may be slow or hit API rate limits (the SKILL.md mentions handling 429 and daily-limit conditions, but expect degraded results if limits are reached); (4) this is third‑party code from an unknown source — run it in a controlled environment or inspect/run tests locally if you need extra assurance. If you want, I can scan the remaining truncated files (especially weather.py) for exact HTTP endpoints, rate-limiting logic, or any unexpected network destinations.

功能分析

Type: OpenClaw Skill Name: go-stargazing Version: 1.4.2 The 'go-stargazing' skill bundle is a legitimate and well-structured tool designed to provide stargazing recommendations based on weather data. It utilizes the Open-Meteo API (api.open-meteo.com) to fetch meteorological parameters such as cloud cover, wind speed, and humidity, and performs complex astronomical calculations (Julian Day, moon phases, and twilight windows) using pure Python logic in 'astronomy.py'. The code follows standard practices, including rate limiting and error handling in 'weather.py', and geographic clustering in 'geo.py'. No evidence of data exfiltration, malicious execution (e.g., eval/exec), or harmful prompt injection was found in the scripts or the SKILL.md instructions.

能力标签

crypto

能力评估

✓ Purpose & Capability

Name/description match the contained code and runtime instructions. The code bundle implements region-level astronomy, geospatial aggregation, scoring and Open‑Meteo-based weather fetches as described. Declared runtime needs (python3, lunarcalendar, shapely, Open‑Meteo) are appropriate for the task; no unrelated credentials or binaries are requested.

✓ Instruction Scope

SKILL.md limits scope to region-level recommendations, describes coarse/finetune phases, and explicitly requires using real Open‑Meteo HTTP API for true-weather mode and to inform users on failure. The docs do not instruct reading unrelated local files or exfiltrating secrets. The code files implement weather fetching, geography, scoring and output drafting consistent with SKILL.md.

✓ Install Mechanism

No install spec is provided (instruction-only install), and the package is shipped as code files plus a small requirements.txt. Dependencies are standard Python packages; nothing is downloaded from arbitrary or obfuscated URLs in the manifest. Note: shapely may require system-level wheel/libs when pip installing.

✓ Credentials

The skill requests no environment variables or credentials. Its external access needs (outbound HTTP to Open‑Meteo) are justified by the stated purpose. There are no requests for unrelated secrets or config paths.

✓ Persistence & Privilege

The skill does not declare always:true and does not request persistent system privileges. It will execute locally when invoked and may make outbound requests; no evidence it modifies other skills or system-wide agent settings.

版本历史

v1.4.2

Tighten skill copy and streamline Go Stargazing description for ClawHub display.

v1.4.1

Refine ECMWF-only national scan packaging and publish current go-stargazing skill.

v1.4.0

Tighten skill metadata, improve trigger description, and remove go-stargazing-trip routing guidance.

v1.3.9

Refine nationwide coarse-screening, scoring guidance, and reply constraints for go-stargazing.

v1.3.8

SKILL.md: 粗筛改为 GFS only 描述同步；去掉多天规则中 go-stargazing-trip 引用；去掉导出边界整节

v1.3.7

粗筛改为 GFS only 大幅减少 national API 调用；SKILL.md 同步更新模型策略描述；细筛阶段保留多模型对比

v1.3.6

粗筛改为单模型 GFS only，大幅减少 national 场景 API 调用量；多模型对比保留至 stage2 细筛阶段

v1.3.5

Documentation cleanup release: removed legacy scoring wording, renamed reply guidance heading to clearer Chinese wording, and kept existing dependency/install declarations intact.

v1.3.4

Dependency declaration release: explicitly added shapely alongside lunarcalendar, included requirements.txt, and clarified install/runtime prerequisites for Python-based real-weather execution.

v1.3.3

Metadata and packaging clarity release: added explicit python3 + lunarcalendar requirements, documented runtime prerequisites for real-weather mode, and kept prior package cleanup plus fetch-limit handling improvements.

v1.3.2

Package cleanup release: removed unused reference/test artifacts and stale packaged files, tightened ignore rules, and kept recent fetch-limit handling improvements including daily-limit exceeded user messaging.

v1.3.1

National sampling refined to ~200 points with hole-filling, representative compensation, and latitude-aware distribution; removed built-in Excel export; improved real-weather fetch handling with model-specific backoff, separate national vs review fetch modes, and daily-limit-exceeded user messaging.

v1.3.0

Default model flow is now GFS + ECMWF with ICON auto recheck on dispute; delivery now returns text recommendation first and Excel as the primary attachment/export; removed professional markdown/json as main delivery path; Excel candidate output merged into a single table with one final score for users.

v1.2.1

Slimmed package by removing heavy reference/geometry assets and keeping lite regional data; preserved lightweight nationwide stargazing recommendation flow.

v1.2.0

Split responsibilities: keep go-stargazing focused on nightly/per-night recommendations only. Multi-day inputs now return nightly-independent results with full metrics and candidate lists; route continuity planning moved to go-stargazing-trip.

v1.1.4

Add dew_point/precip/cloud_base/weather_code to hourly params; enrich trip mode final_reply_draft with full weather metrics and top-N candidates per night

v1.1.3

Add canonical output constraint: callers must use final_reply_draft directly, not re-summarize JSON fields.

v1.1.2

Restore full acknowledgements text in SKILL.md; keep 1.1.1 packaging and behavior fixes unchanged.

v1.1.1

Fix packaging consistency for distribution: replace .gz+exec structure with plain modular Python files, move bundled boundary data to publishable text polygon resources, and verify key recommendation behavior matches the previous release.

v1.1.0

Restore old-style response flow; split astronomical night window vs best shooting window; keep national coarse filter cloud-only; restore refine prompts and multi-day continuity judgement by default; fix moon phase reference drift; slim publish payload for ClawHub compatibility.

元数据

Slug go-stargazing

版本 1.4.2

许可证 MIT-0

累计安装 0

当前安装数 0

历史版本数 29

常见问题

Go Stargazing 是什么？

轻量版区域级拍星推荐技能，适用于今晚或指定日期的全国、省市级拍星条件筛选。默认全国扫描，先按云量粗筛，再结合真实天气细筛，输出逐晚推荐和完整气象信息。它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 332 次。

如何安装 Go Stargazing？

在 OpenClaw 或 Claude Code 对话框中运行命令「/install go-stargazing」即可一键安装，无需额外配置。

Go Stargazing 是免费的吗？

是的，Go Stargazing 完全免费，采用 MIT-0 许可证，可自由下载、安装和使用。

Go Stargazing 支持哪些平台？

Go Stargazing 跨平台运行，可在任意部署了 OpenClaw / Claude Code 的环境中使用（cross-platform）。

谁开发了 Go Stargazing？

由 clhwbd（@clhwbd）开发并维护，当前版本 v1.4.2。

Go Stargazing