jetskibayConnect
← All resources

SKILL / AGENT SKILLS

Inspired visual search skill.

A single-file agent skill: when to reach for Inspired, which focus domain to pick, how to poll without burning the key's allowance, and the words to use when reporting matches.

Mock studioPublic information
Inspired visual search skill
OverviewInstalling itExamples & promptsFormats & downloads

Installing and checking it

Install

Download SKILL.md and put it at inspired-visual-search/SKILL.md inside a skills directory your agent reads:

| Agent | Location | | --- | --- | | Claude Code, one repository | .agents/skills/inspired-visual-search/SKILL.md | | Claude Code, every project | ~/.claude/skills/inspired-visual-search/SKILL.md | | Anything else | Whatever directory that runtime scans for skills |

The directory name matters: it is the skill's identity, and the frontmatter name is inspired-visual-search to match. Start a new session afterwards — most runtimes scan for skills once, at startup, so a skill added mid-session is not discovered.

Nothing installs itself. There is no script, no post-install step and no network call in the skill; it is instructions for a model to read.

Give it something to call

The skill drives Inspired, it does not contain it. Provide one of:

  • The MCP connection, https://inspired.jetskibay.com/mcp, with your API key as a bearer token. The skill expects the tool names list_focus_domains, start_visual_search and get_visual_search.
  • HTTP access plus the key in the environment, if your agent shells out with curl.

Without either, a correctly working skill refuses the task and points at the docs. That is the intended behaviour, not a failure.

Check it took

Ask for something the skill should recognise, with no photo attached:

Where could I buy the lamp in a photo of my living room?

You should see it name the furniture domain rather than the default, mention that a read takes minutes and costs one request, and ask for the image — not start guessing at lamps.

Then try it with a real photo and watch for three things:

  • One start_visual_search call, not several.
  • Polling that waits the interval the response asks for.
  • A report that says spotted and visually similar, never identified or found.

If you see two starts for one image, the agent is not following the skill and you have just spent two of ten requests.

Uninstalling

Delete the directory. Nothing else was written anywhere.

Open Markdown source

Published by Inspired. Fictional brand, working prototype.