From skillrush-town
淘金小镇 | Skillrush Town — 每日 ClawHub Top100 淘金雷达。消费侧:用户问 ClawHub 新 skill、榜单、潜力 skill 时,读取 data/latest.json 总结 Top10、潜力 Skill 与新进榜。维护侧:building, maintaining, forking, or publishing Skillrush Town; tracking ClawHub Top100 downloads snapshots; generating daily Skill reports.
How this skill is triggered — by the user, by Claude, or both
Slash command
/skillrush-town:skillrush-townThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
当用户问"今天有什么新 skill / 什么在涨 / 有什么值得装"这类问题时,不要走维护流程,直接读数据回答:
当用户问"今天有什么新 skill / 什么在涨 / 有什么值得装"这类问题时,不要走维护流程,直接读数据回答:
data/latest.json;不在仓库内时读
https://learnprompt.github.io/skillrush-town/data/latest.json。
回看历史用 data/snapshots/<YYYY-MM-DD>.json,可用日期列表在 data/dates.json。items 前 10 条)、潜力 Skill(报告
data/reports/<date>.md 的「潜力 Skill」节,或按 SKILL.md 潜力标准从 items 现算)、
新进榜(有历史时 prev_rank 为 null 的条目)。https://learnprompt.github.io/skillrush-town/?date=<snapshot_date> 方便用户自己看。limitations 与 comparison_basis.note:缺历史切片时不要把数字说成日环比。When this skill triggers inside the repo, read:
README.md for product positioning and user paths.scripts/clawhub_daily.py before changing data generation.data/dates.json and the latest data/snapshots/*.json before changing the UI.assets/app.js, assets/styles.css, and index.html before changing the page.references/source-contract.md before changing ClawHub request semantics.references/source-adapter-pattern.md before adding changelog, model leaderboard,
or any non-ClawHub monitoring source.references/publishing.md before changing GitHub Pages or Actions.Maintain three surfaces:
Keep the default story simple: "每天从公开榜单和更新日志里淘出值得看的 AI 变化." ClawHub Top100 is the first source because it is complex enough to prove the workflow: runtime request, Convex path, cursor pagination, Top100 normalization, historical diffs, reports, and a public board.
Do not position this as "just a webpage". The page is the town board; the Skill is the workflow contract that lets future agents maintain and extend the town.
Good future source examples:
Installing this Skill only installs the workflow contract; it must not create hidden scheduled jobs during installation.
For a one-off check, the user can say:
Use skillrush-town to read latest.json and summarize today's ClawHub Top10 and potential Skills.
For a daily reminder in Hermes, create a cron job only when the user asks for it. A good default is 10:00 Asia/Shanghai, after the GitHub Actions update window:
Every day, read https://learnprompt.github.io/skillrush-town/data/latest.json,
summarize the snapshot date, Top10, potential Skills, limitations, and link to
https://learnprompt.github.io/skillrush-town/?date=<snapshot_date>.
Codex and Claude Code are usually task runners, not persistent reminder daemons. For them, keep the repo forkable and use GitHub Actions, system cron, or Hermes cron for notifications.
api/query, path
skills:listPublicPageV4.sort=downloads, dir=desc,
nonSuspiciousOnly=true, highlightedOnly=false, numItems=25.nextCursor for 4 pages.GET /api/v1/skills as the primary ranking basis.Before adding a changelog, model leaderboard, release feed, or any non-ClawHub
source, read references/source-adapter-pattern.md and create a source-specific
contract file:
skills/skillrush-town/references/source-contract-<source>.md
Do not merge a new source adapter unless it has:
Every run must produce:
data/snapshots/YYYY-MM-DD.jsondata/reports/YYYY-MM-DD.mddata/latest.jsondata/dates.jsonReports must include new entries, dropped entries, Top10 changes, downloads
growth Top10, stars growth Top10, and potential Skills. If there is no potential
Skill, explicitly write 今日无新增潜力skill.
Never describe the first run, migration run, or missing-history comparison as a strict daily delta.
Include at most 10. A Skill qualifies if any condition is true:
Each potential Skill needs name, rank change, download/star delta, and one short recommendation.
For public README and release text, keep the town/gold-rush metaphor but avoid AI-marketing boilerplate. Prefer concrete use cases over abstract claims. Do not say "empower", "ecosystem flywheel", "comprehensive platform", or similar filler.
These checks must work without Chrome, Playwright, Puppeteer, Camofox, Selenium, or browser login state:
python -m py_compile scripts/clawhub_daily.py
python -m pytest -q
python "${CODEX_HOME:-$HOME/.codex}/skills/.system/skill-creator/scripts/quick_validate.py" skills/skillrush-town
For a live ingestion check, write into a temporary data directory instead of mutating committed data:
TMP=$(mktemp -d)
python scripts/clawhub_daily.py --date 2026-05-04 --data-dir "$TMP/data"
Only run this when a browser is available:
python3 -m http.server 8093
Open http://127.0.0.1:8093/?date=2026-05-04 and verify the date selector,
Top10, limitation panel, potential Skill section, search, and Top100 table.
npx claudepluginhub learnprompt/skillrush-town --plugin skillrush-townSearches local, marketplace, GitHub, and web skill sources to avoid duplicating existing skills. Indexed results are ranked and presented with fork/extend options.
Searches 9 skill registries including SkillsMP, GitHub, and Anthropic Skills for agent skills. Scores by quality/security, paginates results up to 5, and installs locally/globally. Use on 'fetch/install/search skill' or 'MISSING EXTERNAL SKILL'.
Discovers, recommends, and installs skills from verified-skill.com registry. Guides users by tech/domain/project, lists/manages installed skills with npx vskill.