بنقرة واحدة
luxonis-troubleshoot
Diagnose broken, slow, or confusing Luxonis/OAK apps and device setups.
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
القائمة
Diagnose broken, slow, or confusing Luxonis/OAK apps and device setups.
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
استنادا إلى تصنيف SOC المهني
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.
Choose the right Luxonis skill for your current situation.
| name | luxonis-troubleshoot |
| description | Diagnose broken, slow, or confusing Luxonis/OAK apps and device setups. |
| disable-model-invocation | true |
| argument-hint | symptom, error/log, or app path |
| metadata | {"author":"luxonis","version":"1.0.0","status":"stable"} |
Diagnose a broken, slow, or confusing Luxonis/OAK app or device. This is the general
"something is wrong" entry: never refuse, triage, and either fix it or hand off a clean
support packet. Be self-sufficient -- do not assume any other Luxonis skill is installed.
Reference docs and support@luxonis.com, not sibling skills.
End in exactly one state:
SUPPORT_PACKET.md written with the likely
owner and next action.Change one thing, rerun the same reproduction, observe. Never stack changes. Never declare fixed without a passing rerun.
Read the working directory and the user's input. If present, read DEVICE.md and
PROJECT_BRIEF.md. Do not re-interrogate the user about facts a prior session already
captured.
DEVICE.md -- a hint of a previously verified device (model/family/connection/id), not
ground truth. Use it as a starting point, but trust live discovery and debug logs over it:
DHCP, re-cabling, or re-adoption can make it stale, and when the record disagrees with what
the device actually reports, believe the live device and note the drift.PROJECT_BRIEF.md -- tells you what the app is supposed to do; sharpens "wrong output".AGENTS.md/CLAUDE.md as diagnostic source; they are agent guidance.Pin the symptom to a concrete, rerunnable observation. This observation is the definition of fixed.
DEPTHAI_LEVEL=debug (escalate
to trace) to surface XLink/connection/pipeline detail.Use the smallest reliable repro.
One cheap check decides which layer the bug is on: can a minimal known-good DepthAI snippet open the device and stream one frame?
Live DepthAI discovery is the authority here. oakctl list is non-authoritative: RVC2 is
DepthAI-first and never appears in oakctl at all, so an empty oakctl list is not evidence a
device is missing -- confirm with DepthAI discovery before concluding a device is unavailable.
/luxonis-device-setup if the user has it installed -- otherwise continue here. A
hardware fault (orange LED, boot failure, suspected calibration) is a hard stop ->
support-packet to support@luxonis.com.Never reason about the DepthAI API from memory. Confirm every API claim against a
known-good example or the docs, and confirm the depthai version before touching
version-sensitive code -- always DepthAI v3, never v2.
For the app-layer diff, use ~/.luxonis/agent-context/oak-examples as the example reference --
only when a comparison actually helps, not routinely. Source it the same safe way as a build:
~/.luxonis/agent-context (in the user's home) if it does not exist.https://github.com/luxonis/oak-examples over HTTPS (never SSH), shallow and
pinned to main (which tracks DepthAI v3), with git's low-speed guard
(http.lowSpeedLimit=1000, http.lowSpeedTime=60) so a stalled network aborts instead of
hanging. Clone into a temporary sibling path (e.g. oak-examples.tmp) and rename it into
place only on success, using commands appropriate to the host shell, so a failed clone never
leaves a partial checkout behind. If the checkout already exists, refresh it with a
fast-forward-only pull instead.INDEX.md. If the clone fails or is incomplete, do
the diff from live docs instead; never diff against a partial checkout or an older/v2 example.Change exactly one thing, rerun the repro, observe. Add a print/log only if the repro does not already show the answer.
Use this only to pick the right repro and the right page fast. The fixes live in the docs.
| Symptom | Layer | Docs |
|---|---|---|
| Device not found / discovery | device | Device Information; OAK4 / USB getting-started |
| Linux USB permissions, PoE / network / LED | device | USB / OAK4 getting-started; troubleshooting index |
| Device bricked / won't boot | device -> support | my-device-is-bricked |
| Pipeline crash / error | app | diff vs known-good example |
| Bad stereo / depth | app | StereoDepth node |
| Low FPS / latency | app | my-app-is-slow; optimizing |
| Neural network issues | app | neural-network-issues |
| V2 / V3 confusion | version | v2-vs-v3 |
pip or project-dependency changes -> confirm first.
Non-privileged diagnosis (isolated venv, reading logs, rerunning the repro) runs freely.Write SUPPORT_PACKET.md in the current working directory only when local fixing is
exhausted, then tell the user to send it to support@luxonis.com.
# Luxonis Support Packet: <one-line symptom>
## Symptom
What's wrong + the concrete repro observation (command output / measured FPS / frame).
## Environment
OS, Python, depthai version, oakctl version if used, connection (USB/PoE),
device model/family/id (from DEVICE.md if present).
## Device state
Discovered? Verified working before? LED state. Managed vs factory.
## What I tried
One line per hypothesis: change -> rerun result.
## Logs / errors
Smallest relevant excerpt (include DEPTHAI_LEVEL=debug output if relevant).
## Likely owner / next action
Best-guess category (device / network / app / depth / performance / version) + next step.
depthai and oakctl versions are mandatory so support can spot V2/V3 issues immediately.
oakctl is expected to gain a support-file/bundle command; once it exists, prefer generating
that over the ad-hoc levers above.