Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Add new skills to an existing workflow and update all related documentation. Use when user wants to add skills from GitHub URLs to a workflow (e.g., "add this skill to the workflow", "为工作流添加技能"). Triggers on adding skills to workflows, updating workflow documentation after skill additions.
.claude/skills/nicepkg-add-new-skills-to-workflow/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | -5% | 0% |
| case-14 | ✗→✓ | ▲ Improved | -3% | 0% |
| case-06 | ✗→✓ | ▲ Improved | 40% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -20% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 40% | 0% |
Add skills from GitHub to an existing workflow and update all related documentation.
Use skill-downloader to download skills from GitHub:
bashpython .claude/skills/skill-downloader/scripts/download_from_github.py <repo-url> <skill-path> --output <workflow-path>/.claude/skills/
Parse GitHub URL:
https://github.com/user/repo/blob/main/path/to/skill → repo: https://github.com/user/repo, skill-path: path/to/skillhttps://github.com/user/repo/tree/main/.claude/skills/my-skill → repo: https://github.com/user/repo, skill-path: .claude/skills/my-skillExample:
bash# For URL: https://github.com/XIYO/zheon/blob/main/.claude/skills/slidev python .claude/skills/skill-downloader/scripts/download_from_github.py https://github.com/XIYO/zheon .claude/skills/slidev --output ./workflows/talk-to-slidev-workflow/.claude/skills/ # Use --force to overwrite existing python .claude/skills/skill-downloader/scripts/download_from_github.py <repo> <path> --output <target> --force
Read the downloaded SKILL.md to understand:
Add the new skill entry to workflows/<name>/.claude/skill-source.json:
json{ "skill-name": { "source": "https://github.com/user/repo", "path": "path/to/skill" } }
Update these files (all that exist for the workflow):
| File | Updates Required | |------|------------------| | workflows/<name>/.claude/skill-source.json | Add new skill source entry | | workflows/<name>/README.md | Skill count, skill table, pipeline | | workflows/<name>/README_cn.md | Same as above (Chinese) | | workflows/<name>/AGENTS.md | Available skills list, recommended sequences | | website/content/en/workflows/<name>.mdx | Skill count, skill table, pipeline | | website/content/zh/workflows/<name>.mdx | Same as above (Chinese) | | README.md (root) | Skill count in workflow table | | README_cn.md (root) | Same as above (Chinese) |
markdown | skill-name | Brief description of what it does |
Stage X: Category ├── existing-skill → Description └── new-skill → Description
Given: Add slidev and slidev-presentations skills
Step 1: Download
bashpython .claude/skills/skill-downloader/scripts/download_from_github.py https://github.com/XIYO/zheon .claude/skills/slidev --output ./workflows/talk-to-slidev-workflow/.claude/skills/ python .claude/skills/skill-downloader/scripts/download_from_github.py https://github.com/clearfunction/cf-devtools skills/slidev-presentations --output ./workflows/talk-to-slidev-workflow/.claude/skills/
Step 2: Read downloaded skills to understand their purpose
Step 3: Update skill-source.json with new skill entries
Step 4: Update all 8 files:
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 12,320 | 11,758 | -5% | 1 | 1 | 0% | 2,336 | 1,725 | -26% | 0 | 0 | — |
case-02 | fail→fail | 11,207 | 7,150 | -36% | 1 | 1 | 0% | 2,070 | 1,607 | -22% | 0 | 0 | — |
case-03 | fail→fail | 4,889 | 5,674 | +16% | 1 | 1 | 0% | 279 | 1,522 | +446% | 0 | 0 | — |
case-04 | fail→pass | 11,933 | 3,942 | -67% | 1 | 1 | 0% | 2,010 | 1,908 | -5% | 0 | 0 | — |
case-05 | pass→pass | 4,958 | 1,743 | -65% | 1 | 1 | 0% | 825 | 1,402 | +70% | 0 | 0 | — |
case-14 | fail→pass | 9,218 | 1,953 | -79% | 1 | 1 | 0% | 1,409 | 1,369 | -3% | 0 | 0 | — |
case-06 | fail→pass | 5,693 | 2,499 | -56% | 1 | 1 | 0% | 1,060 | 1,486 | +40% | 0 | 0 | — |
case-07 | fail→pass | 11,864 | 2,945 | -75% | 1 | 1 | 0% | 2,025 | 1,628 | -20% | 0 | 0 | — |
case-08 | pass→pass | 4,646 | 5,221 | +12% | 1 | 1 | 0% | 766 | 2,026 | +164% | 0 | 0 | — |
case-09 | fail→pass | 8,071 | 4,300 | -47% | 1 | 1 | 0% | 1,281 | 1,798 | +40% | 0 | 0 | — |
case-10 | pass→pass | 9,283 | 4,557 | -51% | 1 | 1 | 0% | 1,616 | 1,824 | +13% | 0 | 0 | — |
case-11 | fail→pass | 3,851 | 2,860 | -26% | 1 | 1 | 0% | 561 | 1,576 | +181% | 0 | 0 | — |
case-12 | fail→pass | 12,343 | 3,074 | -75% | 1 | 1 | 0% | 2,185 | 1,582 | -28% | 0 | 0 | — |
case-13 | fail→pass | 6,779 | 2,186 | -68% | 1 | 1 | 0% | 964 | 1,355 | +41% | 0 | 0 | — |
case-15 | pass→pass | 9,204 | 3,174 | -66% | 1 | 1 | 0% | 1,347 | 1,567 | +16% | 0 | 0 | — |
case-16 | fail→fail | 10,994 | 2,489 | -77% | 1 | 1 | 0% | 1,640 | 1,480 | -10% | 0 | 0 | — |
case-17 | fail→pass | 7,385 | 2,692 | -64% | 1 | 1 | 0% | 1,137 | 1,551 | +36% | 0 | 0 | — |
case-18 | pass→pass | 5,465 | 2,867 | -48% | 1 | 1 | 0% | 1,102 | 1,594 | +45% | 0 | 0 | — |
case-19 | fail→fail | 12,705 | 2,540 | -80% | 1 | 1 | 0% | 2,258 | 1,531 | -32% | 0 | 0 | — |
case-20 | pass→pass | 3,808 | 3,761 | -1% | 1 | 1 | 0% | 691 | 1,836 | +166% | 0 | 0 | — |
case-21 | pass→pass | 1,551 | 2,330 | +50% | 1 | 1 | 0% | 224 | 1,497 | +568% | 0 | 0 | — |
case-22 | fail→pass | 6,454 | 2,591 | -60% | 1 | 1 | 0% | 1,127 | 1,517 | +35% | 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. 22 cases were attempted, and 19 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 +45 percentage points is the difference between those two pass rates over the 19 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.