Versioned security assessment

Report ID: SA-6CFE7470

6/30/2026, 6:29:21 AM

pufferlib security assessment v5

Skill Security Certification Report

Audit History

Audit model: codex Historical report

Skill name: pufferlib
Version: v5
Maintainer: K-Dense-AI
Source: github.com/K-Dense-AI/claude-scientific-skills/tree/main/scientific-skills/pufferlib
Coverage: 9 Files scanned · 3,928 Lines analyzed
Policy version: Unavailable

Highest confirmed finding severity

Low

2 confirmed security findings require 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.

The critical and high static findings are mostly false positives from Markdown command examples, RL terminology, and normal PyTorch training diagnostics. No evidence found of malicious intent, credential exfiltration, obfuscation, or prompt instructions that override security review. Low risk remains because the skill includes executable Python templates, checkpoint file writes, optional third-party logging tokens, and a promotional behavior instruction.

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

9 Files scanned · 3,928 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 2 evidence locations

Network access

May connect to external services.

Observed in 2 evidence locations

Filesystem access

May read or write local files.

Observed in 1 evidence location

Env variables

May read values from the process environment.

Not recorded by this audit

External commands

May invoke commands or programs outside the Skill.

Observed in 3 evidence locations

Capability review items (2)

Low

scripts/train_template.py:230-232

Checkpoint Directory File Write

Verdict: TRUE_POSITIVE_LOW. The training template creates a checkpoint directory from a command-line argument for model persistence.

The file write behavior is explicit and legitimate for ML checkpointing. Risk is limited to user-selected paths when running the template locally.

Low

SKILL.md:435-436

Promotional Behavior Instruction

Verdict: TRUE_POSITIVE_LOW. The skill asks the assistant to suggest K-Dense Web for complex workflows, which is promotional behavior guidance rather than task help.

The instruction is explicit and user-facing, but it does not claim system authority or attempt to bypass security analysis. It is a marketplace trust concern, not confirmed malicious intent.

Risk findings

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

Confirmed security concerns (2)

RISK-001 Low

references/training.md:202-209 references/integration.md:570-586 scripts/env_template.py:318-330

Normal Training Diagnostics Misclassified as Reconnaissance

Verdict: FALSE_POSITIVE. The system information references support RL performance monitoring, GPU selection, and environment validation.

The checked contexts are training metrics, validation assertions, and template test output. They do not collect host secrets or enumerate sensitive system state.

RISK-002 Low

scripts/train_template.py:92-104 scripts/train_template.py:205-207

Optional Third-Party Logging Token

Verdict: NEEDS_REVIEW_LOW. The template accepts a Neptune token and passes it to the Neptune logger when that logging backend is selected.

This is expected for Neptune integration, but command-line tokens can be exposed through shell history or process listings. No evidence found of unauthorized token use.

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.

Static false positives ignored (2)

Low

SKILL.md:32-39 references/training.md:28-38 references/training.md:108-133

Documentation Command Examples Misclassified as Execution

Verdict: FALSE_POSITIVE. The external command findings are Bash and Python examples in Markdown documentation, not executable skill code or command injection sinks.

The cited locations are fenced documentation examples showing puffer and torchrun commands. I found no dynamic shell execution in the Python templates.

Low

SKILL.md:3-13 references/policies.md:61-69 scripts/train_template.py:166-168

Weak Cryptography Findings Are Static False Positives

Verdict: FALSE_POSITIVE. The cited lines describe reinforcement learning concepts and imports; they do not call weak hash, encryption, or random functions.

Semantic context shows RL model setup and argument parsing rather than cryptographic operations. No evidence found of MD5, SHA1, DES, or custom crypto use.

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