Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →使用 agent-browser 帮用户将内容发到社交媒体上。当用户需要发布内容、推送文章、上传文章、发帖到社交平台时使用此 skill。
.claude/skills/jihe520-social-push/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-06 | ✗→✓ | ▲ Improved | -47% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -35% | 0% |
| case-09 | ✗→✓ | ▲ Improved | -18% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -10% | 0% |
| case-11 | ✗→✓ | ▲ Improved | -6% | 0% |
用户输入 $ARGUMENTS
你需要使用 bash 运行 agent-browser，并参考 references 中对应平台的 workflow，帮助用户将文章、图片上传到对应的社交平台上
open -na "Google Chrome" / "Microsoft Edge" --args --remote-debugging-port=9222, 确保用户打开的浏览器支持远程调试agent-browser --auto-connect 自动连接用户的浏览器agent-browser snapshot -i 确认元素 ref，因为页面状态变化可能导致 ref 编号变化agent-browser --help 可用命令网页交互可能发生变化，references 下面的 workflow 可能失效，按以下步骤修复：
agent-browser snapshot 查看当前页面的详细元素agent-browser eval "js" 查看具体 html 元素当用户询问需要新添加一个平台时候，按以下步骤添加：
agent-browser --help 查看可用命令 和 agent-browser 的 skill小红书图文 ：查看小红书图文发布简短文章图文时候需要的 workflow小红书长文 ：查看小红书长文用户发送长文本时候需要的 workflowX推文 ：查看X推文发布推文时候需要的 workflow微博 ：查看微博发布微博时候需要的 workflow微信公众号文章 ：查看微信公众号文章发布公众号文章时候需要的 workflow掘金文章 ：查看掘金文章发布掘金文章并自动保存草稿的 workflowLinuxDo发帖 ：查看LinuxDo发帖发布帖子（含类别与标签选择）的 workflow| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-04 | fail→fail | 2,235 | 3,402 | +52% | 1 | 1 | 0% | 301 | 1,148 | +281% | 0 | 0 | — |
case-01 | fail→fail | 3,933 | 12,109 | +208% | 1 | 1 | 0% | 521 | 1,608 | +209% | 0 | 0 | — |
case-02 | fail→fail | 14,857 | 9,469 | -36% | 1 | 1 | 0% | 2,601 | 1,175 | -55% | 0 | 0 | — |
case-03 | fail→fail | 10,664 | 9,545 | -10% | 1 | 1 | 0% | 1,301 | 1,340 | +3% | 0 | 0 | — |
case-05 | pass→pass | 11,178 | 3,989 | -64% | 1 | 1 | 0% | 1,967 | 1,410 | -28% | 0 | 0 | — |
case-06 | fail→pass | 11,678 | 2,891 | -75% | 1 | 1 | 0% | 2,038 | 1,074 | -47% | 0 | 0 | — |
case-07 | fail→pass | 16,751 | 5,633 | -66% | 1 | 1 | 0% | 2,418 | 1,564 | -35% | 0 | 0 | — |
case-08 | fail→fail | 9,993 | 3,150 | -68% | 1 | 1 | 0% | 1,557 | 1,184 | -24% | 0 | 0 | — |
case-09 | fail→pass | 8,838 | 2,947 | -67% | 1 | 1 | 0% | 1,437 | 1,172 | -18% | 0 | 0 | — |
case-10 | fail→pass | 10,824 | 5,022 | -54% | 1 | 1 | 0% | 1,703 | 1,537 | -10% | 0 | 0 | — |
case-11 | fail→pass | 7,204 | 2,677 | -63% | 1 | 1 | 0% | 1,127 | 1,064 | -6% | 0 | 0 | — |
case-12 | fail→pass | 13,275 | 3,664 | -72% | 1 | 1 | 0% | 1,927 | 1,199 | -38% | 0 | 0 | — |
case-13 | pass→pass | 9,266 | 2,205 | -76% | 1 | 1 | 0% | 1,433 | 985 | -31% | 0 | 0 | — |
case-14 | pass→pass | 6,956 | 3,236 | -53% | 1 | 1 | 0% | 1,199 | 1,161 | -3% | 0 | 0 | — |
case-15 | fail→pass | 13,082 | 2,258 | -83% | 1 | 1 | 0% | 2,041 | 1,025 | -50% | 0 | 0 | — |
case-16 | pass→pass | 5,242 | 1,627 | -69% | 1 | 1 | 0% | 708 | 906 | +28% | 0 | 0 | — |
case-17 | fail→pass | 13,710 | 3,427 | -75% | 1 | 1 | 0% | 1,886 | 1,191 | -37% | 0 | 0 | — |
case-18 | pass→pass | 5,761 | 6,101 | +6% | 1 | 1 | 0% | 892 | 1,586 | +78% | 0 | 0 | — |
case-19 | fail→fail | 7,890 | 8,147 | +3% | 1 | 1 | 0% | 1,256 | 1,850 | +47% | 0 | 0 | — |
case-20 | pass→pass | 6,031 | 7,625 | +26% | 1 | 1 | 0% | 818 | 1,264 | +55% | 0 | 0 | — |
case-21 | fail→pass | 8,668 | 6,322 | -27% | 1 | 1 | 0% | 1,309 | 1,313 | +0% | 0 | 0 | — |
case-22 | fail→pass | 11,871 | 4,910 | -59% | 1 | 1 | 0% | 1,754 | 1,560 | -11% | 0 | 0 | — |
case-23 | pass→pass | 11,829 | 4,625 | -61% | 1 | 1 | 0% | 1,665 | 1,377 | -17% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 23 cases were attempted, and 20 counted toward the lift figure. The other 3 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +43 percentage points is the difference between those two pass rates over the 20 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.