with one click
luxonis-guide
Choose the right Luxonis skill for your current situation.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Menu
Choose the right Luxonis skill for your current situation.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
Build a thin, runnable Luxonis/OAK proof of concept from a project brief by adapting a known-good example.
Interview the user and produce a buildable Luxonis project brief.
Bring a Luxonis OAK camera to a verified device-ready state.
Diagnose broken, slow, or confusing Luxonis/OAK apps and device setups.
| name | luxonis-guide |
| description | Choose the right Luxonis skill for your current situation. |
| disable-model-invocation | true |
| argument-hint | what are you trying to do? |
| allowed-tools | Bash(ls *) Bash(cat *) Bash(grep *) Bash(echo *) Bash(head *) Bash(true) |
| metadata | {"author":"luxonis","version":"1.0.0","status":"stable"} |
Route the user to one Luxonis skill that fits their current situation. Recommend the next command and stop. Do not do that skill's work here.
End in exactly one state:
Most work runs along one spine, with one side path.
Main flow: idea -> working demo
luxonis-device-setupluxonis-project-interviewluxonis-build-pocSteps 1 and 2 are order-flexible: the interview works with no device yet, so a user waiting on hardware can sharpen the idea first. Step 3 is the convergence point -- it wants both a brief and a verified device.
Side path: something is broken, slow, or confusing -> luxonis-troubleshoot. Branch off
at any point, then come back to the spine.
Preflight the current working directory for flow markers. Existence is the signal; do not deep-parse.
!ls -1 2>/dev/null | grep -E '^(DEVICE|PROJECT_BRIEF)\.md$' || echo "no markers"
!cat PROJECT_BRIEF.md 2>/dev/null | head -3 || true
On agents that support shell injection (e.g. Claude Code) the lines above are filled in
before routing. On agents that do not, ignore them and list the directory / read
PROJECT_BRIEF.md yourself.
Combine the markers with what the user said:
DEVICE.md present -> a device was already verified here.PROJECT_BRIEF.md present -> the idea was already shaped into a brief.Map the situation to one entry:
luxonis-device-setup.DEVICE.md) but no PROJECT_BRIEF.md, and wants to build
something -> luxonis-project-interview.PROJECT_BRIEF.md exists and the user wants it running -> luxonis-build-poc.luxonis-troubleshoot.If the situation is clear, route immediately. If it is ambiguous (no argument, or something vague like "help with my OAK"), ask exactly one clarifying question, then commit:
Do you already have a camera connected and working, or are you starting from just an idea?
Never ask a second question. Never start shaping the idea -- that is
luxonis-project-interview's job, not this router's.
"My camera isn't working" splits two ways. This is the spot where the one clarifying question earns its keep:
luxonis-troubleshoot.support@luxonis.com, not troubleshoot.routed -- name the next command and one line on where it sits:
Run
/luxonis-device-setupnext -- that verifies the camera, then we shape the idea and build.
Plugin installs may namespace the command, e.g. /luxonis:luxonis-device-setup.
out-of-scope -- do not force a route. Name the gap and point onward:
support@luxonis.com.