registry / @camstack/addon-model-studio / 1.0.10

@camstack/addon-model-studio@1.0.10

Custom detection model registry, conversion & distribution for CamStack

AI Security Review

scanned 7d ago · by lpm-firewall-ai

A user-invoked model-conversion action downloads an arbitrary `sourceUrl`. Its Hugging Face authorization guard uses a substring check, enabling credential disclosure to an attacker-controlled URL containing `huggingface.co`.

Static reason

No blocking static signals were detected.

Trigger

Invoke the `convertModel` custom action with a crafted `sourceUrl` while an HF token is configured.

Impact

Exfiltration of the configured Hugging Face token; redirected downloads also permit untrusted model input to reach local conversion libraries.

Mechanism

Arbitrary URL fetch with incorrectly scoped Hugging Face authorization header.

Rationale

No malicious lifecycle or persistence behavior is present, but the token-host validation flaw creates a concrete user-triggered credential-exfiltration path. Downgrade to warn pending hostname allowlisting and redirect validation.

Evidence

package.jsondist/index.jsdist/index.mjsdist/model-studio.addon.mjspython/convert.pypython/convert_lib.py

Network endpoints1

huggingface.co

Decision evidence

public snapshot

AI called this Suspicious at 91.0% confidence with false-positive risk.

@camstack/addon-model-studio@1.0.10

AI Security Review

Decision evidence

Source & flagged code

Source & flagged code

Findings

Source & flagged code