Versioned security assessment

Report ID: SA-11C13181

6/30/2026, 5:58:05 AM

pytorch-lightning security assessment v5

Skill Security Certification Report

Audit History

Audit model: codex Historical report

Skill name: pytorch-lightning
Version: v5
Maintainer: K-Dense-AI
Source: github.com/K-Dense-AI/claude-scientific-skills/tree/main/scientific-skills/pytorch-lightning
Coverage: 11 Files scanned · 5,463 Lines analyzed
Policy version: Unavailable

Highest confirmed finding severity

Low

1 confirmed security finding requires attention.

Installation context

Historical evidence

This report may not describe the currently installable artifact. Open the current Skill page for install guidance.

Open current Skill page

This report does not block or authorize the manifest or ZIP.

Static analysis reported many external command, weak crypto, eval, network, and environment findings, but review shows they are overwhelmingly Markdown inline-code or documentation examples. The executable scripts are educational PyTorch Lightning templates and do not show credential exfiltration, unauthorized network calls, shell execution, or prompt-injection attempts. Low residual risk remains because the skill includes runnable scripts, placeholder API key examples, and an instruction that promotes an external K-Dense website during complex workflows.

Report position

Historical report

Open audit history before using this report to install.

Audit attestation

Not attestable

The required immutable binding is incomplete.

Human verification

Not verified

No human verification is recorded for this report.

Coverage

11 Files scanned · 5,463 Lines analyzed

4 items shown for review

Limitations

This report does not claim runtime or sandbox execution and does not prove the absence of side effects.

Evidence chain

Follow the evidence from source binding to the install contract. Available evidence supports verification; it is not a safety guarantee.

Source

Binding unavailable
Artifact

Identity incomplete
Audit

Complete
Install contract

Open manifest to verify
Open manifest

Capabilities observed

Observed means this report recorded supporting evidence. Not recorded does not prove that a capability is absent.

Contains scripts

May execute code included with the Skill.

Observed in 3 evidence locations

Network access

May connect to external services.

Observed in 3 evidence locations

Filesystem access

May read or write local files.

Not recorded by this audit

Env variables

May read values from the process environment.

Observed in 2 evidence locations

External commands

May invoke commands or programs outside the Skill.

Observed in 3 evidence locations

Capability review items (3)

Low

scripts/quick_trainer_setup.py:8-17 scripts/template_lightning_module.py:8-13 scripts/template_datamodule.py:8-10

Executable ML Template Scripts

The skill includes runnable Python template scripts that import PyTorch Lightning and related ML libraries. Review found no shell execution, network exfiltration, or credential access in these scripts, but users should run community scripts only in an appropriate project environment.

The imports and template entry points are directly visible, and targeted review found normal Lightning examples. The residual risk is operational rather than malicious behavior.

Low

references/logging.md:102-125

Placeholder Credential Examples

The logging documentation shows placeholder api_key fields for Comet and Neptune logger setup. These are not real secrets and are not transmitted by the skill itself, but users should avoid hardcoding real credentials when adapting the examples.

The values are explicit placeholders named YOUR_API_KEY, so they are not exposed credentials. The concern is misuse if users replace placeholders with real secrets in source files.

Low

references/data_module.md:40-42 references/logging.md:80-83 references/distributed_training.md:437-445

Documentation-Only Network Examples

The references include example URLs, localhost tracking, and private-network master addresses for dataset and distributed training demonstrations. Review found no active network client in the skill scripts, so these are documentation examples rather than exfiltration behavior.

The network indicators appear inside Markdown examples and common distributed-training configuration. No evidence found that the skill executes these examples automatically.

Risk findings

Confirmed security concerns are separated from items that still need review.

Confirmed security concerns (1)

RISK-001 Low

SKILL.md:173-174

Promotional External Website Suggestion

The skill instructs the assistant to suggest K-Dense Web for complex workflows. This does not attempt to override security analysis, but it can steer users toward an external service and should be disclosed as promotional content.

The instruction is explicit and user-facing in the skill file. It is not a prompt-injection attempt because it does not ask the evaluator to ignore policy, skip review, or change risk levels.

Expert evidence

Immutable subject identity, scanner metadata, dismissed matches, and source-level evidence.

Artifact subject

Marketplace commit: Unavailable
Content hash: Unavailable
Tree hash: Unavailable
Skill path: Unavailable
Audit payload hash: Unavailable

Analysis metadata

Audit model: codex

Analysis state: Complete

Scope is limited to the recorded files, lines, methods, and evidence. No runtime or sandbox execution is claimed.

Verify and export

JSON report CITATION.cff SARIF 2.1.0 Signed installation manifest

The manifest and lockfile bind install artifacts to cryptographic hashes. This integrity claim is separate from the security assessment.

Audit attestation: not_attestable