training-code-auditor

training-code-auditor is a Model training & evaluation skill for AI agents, published by KeploreAI-Lab in the MindAct repository on GitHub. Skillspace rates training-code-auditor F (45/100) and the safety scan rated it risky; the rating is based on static analysis only.

Risky. This skill has patterns associated with unsafe behaviour. Read the source carefully before running it.

Use this skill BEFORE executing any machine learning training script — especially YOLOv8, PyTorch, or custom training code. This skill audits the generated training code for: (1) correct loss functions for the task (e.g. BCE+DiceLoss for segmentation, NOT MSE), (2) correct evaluation metrics at the right IoU thresholds (mAP@IoU=0.75 for instance segmentation, not just mAP@0.5), (3) fake, placeholder, random, or stub code that would silently produce garbage results, (4) data pipeline correctness (no random labels, real file paths, correct normalization). Trigger this skill whenever training code is generated or reviewed, when the user says "check my training script", "is this code correct", "audit the training code", "review before training", or any time code contains TODO, pass, random, np.random, torch.randn used as fake data, or placeholder comments. NEVER let training start without passing this audit.

Grade

F

Quality

45 / 100

Safety

Risky

GitHub stars

85

Unproven: rated by static analysis only, not yet compared head-to-head.

This rating describes the version of the repository we crawled; it may have changed since.

Skillspace indexes and rates this skill; it does not host or endorse it. Run third-party skills at your own risk.