ML Evolution Agent
PassAudited by ClawScan on May 10, 2026.
Overview
The skill appears to be a purpose-aligned Kaggle/ML template with some provenance and completeness gaps, but no evidence of hidden execution, data exfiltration, or destructive behavior.
This looks safe as a user-directed ML competition template, but verify the missing implementation/dependencies before running it, keep Kaggle auto-submit disabled unless you approve it, and avoid placing secrets or sensitive raw data into the memory files.
Findings (3)
Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.
The skill may not run as advertised without additional, unreviewed code or manual adaptation.
The skill references a package/class and additional files that are not included in the supplied manifest, so the runnable implementation and provenance are incomplete even though no hidden installer is shown.
from ml_evolution import MLEvolutionAgent ... FEATURE_ENGINEERING.md ... MODEL_CONFIGS.md ... EVOLUTION_RULES.md ... train_evolved.py
Treat this as a template unless the missing implementation is supplied and reviewed; verify dependencies and source before running additional code.
If enabled, the script could submit files to a Kaggle competition under the user's account.
The template includes an optional Kaggle CLI submission path, which would use the user's Kaggle account credentials and consume submission quota if uncommented.
# Uncomment to auto-submit # subprocess.run(['kaggle', 'competitions', 'submit', '-c', COMPETITION,
Keep auto-submit commented unless you explicitly approve the file, competition name, and message; review Kaggle credentials and quota before enabling.
Experiment details, model choices, scores, and notes may be reused in later phases.
The skill intentionally maintains persistent experiment memory across phases, which is expected for an evolving ML workflow but should remain scoped to non-sensitive experiment metadata.
**Purpose**: Record every experiment in detail ... **When to update**: After every phase
Avoid storing raw private datasets, secrets, or credentials in the memory files; periodically review or clear memory if using sensitive competition data.