VibeFuse skills are folders of Markdown that the agent loads on demand. Until it decides a skill is relevant, all it sees is the name and the description line. That is why the description is the most important sentence you will write.
One folder in C:\vibefuse\skills, named after the slug you want to type.
Write a description that names the trigger words the user will actually say.
Add any script it needs, then call it with a slash command.
Step 1. The folder
C:\vibefuse\skills\site-health\
SKILL.md required, front matter plus the procedure
scripts\check.ps1 optional, real work it can run
references\thresholds.md optional, loaded only when needed
Step 2. SKILL.md
---
name: site-health
description: Check fuseintelligence.org pages for errors and report status. Use when the user says check the site, is the site up, or run site health.
---
# Site Health
Run scripts/check.ps1 and report a table of URL, HTTP status and byte size.
Rules:
1. Test the landing page, the product page and the docs page.
2. Treat anything that is not 200 as a failure and say so plainly.
3. Quote the raw status code for every URL, never summarise as fine.
4. If a page fails twice, stop and report the blocker instead of retrying blindly.
Step 3. A script it can actually run
# scripts/check.ps1
$urls = @(
'https://fuseintelligence.org/',
'https://fuseintelligence.org/products/vibefuse.php',
'https://fuseintelligence.org/docs/vibefuse.php'
)
foreach ($u in $urls) {
try {
$r = Invoke-WebRequest $u -UseBasicParsing -TimeoutSec 20
"{0} {1} {2} bytes" -f $r.StatusCode, $u, $r.RawContentLength
} catch {
"FAIL $u $($_.Exception.Message)"
}
}
Step 4. Wire it up
Ask Jarvis to install it, or call the tool directly:
create_skill
name: site-health
description: Check fuseintelligence.org pages for errors and report status.
body: full SKILL.md text
files: { 'scripts/check.ps1': '...' }
Then invoke it three ways:
- Slash command - typing /site-health in the Jarvis panel makes the skill mandatory for that turn.
- Natural language - is the site up right now matches the description and loads the skill automatically.
- From another skill - one skill can execute another by slug, which is how you compose pipelines.
Authoring rules that matter
- Name the trigger words. Include the phrasings a human would use, not synonyms from a thesaurus.
- Number the steps. Ordered actions with real commands beat paragraphs of advice.
- Ship the script. A skill that owns working code is dramatically more reliable than one that improvises.
- State the stop condition. Tell the agent when to stop and report instead of looping forever.
- Keep it lean. Long skills dilute attention. Split big workflows into two skills that chain.
Short skills that own a script outperform long skills that own opinions.
Try it now: create a skill called night-shift that writes a morning briefing from your calendar, email and open issues, then run it with /night-shift. Post what you built below and I will review it.