Package, publish, inspect, and troubleshoot MMDetection3D serving utilities.
Scanned 9/8/2026
Install to Claude Code
npx -y skills add VectorSpaceLab/AREX-Skill --skill serving-tools --agent claude-codeInstalls into .claude/skills of the current project.
Are you the author of Serving Tools?
Add the live security badge to your README — it updates automatically with every re-scan.
[](https://www.skillsdirectory.com/skills/vectorspacelab-serving-tools)More formats (shields.io, HTML) on the badges page.
---
name: serving-tools
description: "Package, publish, inspect, and troubleshoot MMDetection3D serving utilities."
disable-model-invocation: true
metadata:
disco-role: operating
license: Apache 2.0
---
# MMDetection3D serving-tools sub-skill
Use this sub-skill when the task is one of:
- package a checkpoint for TorchServe-style serving,
- verify the packaging inputs before export,
- publish or convert a checkpoint for release,
- inspect training logs, FLOPs, or throughput,
- print configs or fuse Conv+BN for utility work.
This sub-skill is safe by default. The bundled checker only validates paths and required artifacts; it does not start servers, build containers, download weights, or run model execution.
## Start here
1. Identify the utility family: serving packaging, checkpoint publishing/conversion, log analysis, FLOPs/throughput, or config inspection.
2. Before packaging, run [`scripts/check_serving_artifacts.py`](scripts/check_serving_artifacts.py) to confirm the config, checkpoint, handler, model name, and export target are ready.
3. For the serving flow and handler assumptions, read [`references/serving.md`](references/serving.md).
4. For the utility script catalog and safe-use boundaries, read [`references/tools-reference.md`](references/tools-reference.md).
5. For common failures and stop conditions, read [`references/troubleshooting.md`](references/troubleshooting.md).
## Route away
- Choosing a model family, config alias, or checkpoint family belongs in `configuration-model-zoo`.
- Raw inference demo commands or inferencer APIs belong in `inference`.
- Training, testing, evaluation, TTA, distributed launch, and Slurm workflows belong in `training-evaluation`.
- Geometry, box conversion, coordinate systems, and visualization internals belong in `structures-visualization`.
- Dataset preparation and conversion belong in `data-preparation`.
- Custom component implementation belongs in `customization-extensions`.
## Minimal operating rules
- Do not run TorchServe, Docker, download checkpoints, or benchmark jobs unless the user explicitly asks for execution.
- Treat the bundled checker as a preflight only; it prints missing artifacts and exits, but it does not repair anything.
- Keep utility advice tied to local files and explicit config/checkpoint pairs.
- Prefer the bundled references over free-form guesses when the user asks about script flags or expected artifacts.
Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.
No comments yet. Be the first to comment!