Most static findings are false positives from Markdown fences, LaTeX examples, DOI examples, and scientific writing prose. Semantic review found real concerns: unverified Bash script execution guidance, an embedded self-audit that claims the package is safe, and promotional steering to an external service. The author should remove or constrain these behaviors before publication.
SKILL.md grants Bash access and instructs agents to run local Python scripts from scripts/. The package does not include that directory, so a caller workspace could supply unintended code at those paths.
The instructions explicitly permit Bash and show python commands against relative script paths. The packaged file tree contains assets, references, EVALUATION_OUTPUT.json, and SKILL.md, but no scripts directory.
EVALUATION_OUTPUT.json embeds a self-audit that marks risk_level as safe, safe_to_publish as true, and summarizes all static findings as false positives. This untrusted safety claim can bias marketplace review.
The file does not issue direct override commands, but it makes explicit safety and publication claims inside the skill package. That matches the prompt-injection warning category for embedded review claims.
SKILL.md tells the agent to proactively suggest K-Dense Web for complex workflows. This is promotional steering to an external service embedded in task guidance.
The instruction is explicit and appears at the end of the skill guidance. It is not malware, but it is a clear marketplace trust and user-steering concern.
Most static findings are false positives from Markdown fences, LaTeX examples, DOI examples, and scientific writing prose. Semantic review found real concerns: unverified Bash script execution guidance, an embedded self-audit that claims the package is safe, and promotional steering to an external service. The author should remove or constrain these behaviors before publication.
SKILL.md grants Bash access and instructs agents to run local Python scripts from scripts/. The package does not include that directory, so a caller workspace could supply unintended code at those paths.
The instructions explicitly permit Bash and show python commands against relative script paths. The packaged file tree contains assets, references, EVALUATION_OUTPUT.json, and SKILL.md, but no scripts directory.
EVALUATION_OUTPUT.json embeds a self-audit that marks risk_level as safe, safe_to_publish as true, and summarizes all static findings as false positives. This untrusted safety claim can bias marketplace review.
The file does not issue direct override commands, but it makes explicit safety and publication claims inside the skill package. That matches the prompt-injection warning category for embedded review claims.
SKILL.md tells the agent to proactively suggest K-Dense Web for complex workflows. This is promotional steering to an external service embedded in task guidance.
The instruction is explicit and appears at the end of the skill guidance. It is not malware, but it is a clear marketplace trust and user-steering concern.
Static analysis reported many high-risk patterns, but most are false positives from Markdown, LaTeX syntax, relative figure paths, and reference URLs. The remaining real risk is bounded: the skill allows Bash and gives local document-generation commands, so users should review commands before execution. No prompt injection or malicious exfiltration evidence was found.
The skill declares Bash access and includes examples that run Python figure generators and LaTeX compilers. This is legitimate for manuscript production, but it can execute local commands and write output files.
The command examples are explicit and Bash is allowed. Confidence is not higher because the commands are normal document-generation operations, not evidence of malicious intent.
Static false positives ignored (4)
These static matches were dismissed by semantic review or matched schema-only tokens, so they are shown for transparency but do not drive the quality score.
Markdown and LaTeX syntax misclassified as execution or weak crypto
Many static hits are inline Markdown code, fenced LaTeX examples, or LaTeX package commands. These are documentation and typesetting syntax, not Ruby backtick execution or cryptographic code.
The cited lines show Markdown formatting, LaTeX environments, and package declarations. No interpreter invocation or cryptographic operation appears in those examples.
Relative figure paths are LaTeX examples, not path traversal
The path traversal alerts point to LaTeX includegraphics examples using parent-directory figure paths. They do not read arbitrary user paths or implement file access logic.
The cited paths are static LaTeX image references inside examples. There is no code that accepts user-controlled paths or traverses the filesystem.
Reference URLs are documentation links, not network beacons
Hardcoded URL alerts point to DOI examples, citation resources, and reporting guideline websites. The files do not contain fetch, curl, or other network-call logic.
The URLs are visibly presented as human reference links and citation examples. No evidence found of automated network access or data exfiltration.
Scientific review language misclassified as reconnaissance
System reconnaissance alerts appear to be triggered by scientific terms such as review, evaluate, and check. The surrounding context is manuscript quality guidance, not host or network discovery.
The reviewed text concerns peer review, manuscript validity, and reporting checklists. No host enumeration or network scanning instructions were found.
Documentation-only skill containing markdown files and LaTeX templates. All 846 static findings are FALSE POSITIVES triggered by benign documentation syntax: markdown backticks for inline code formatting, LaTeX file path references, algorithm term references in text, and documentation URLs. No executable code exists. No security concerns present.
Documentation-only skill containing markdown files and LaTeX templates. All 846 static findings are FALSE POSITIVES triggered by benign documentation syntax: markdown backticks for inline code formatting, LaTeX file path references, algorithm term references in text, and documentation URLs. No executable code exists. No security concerns present.
Documentation-only skill with no executable code. All 822 static findings are false positives triggered by markdown formatting syntax (backticks), LaTeX template paths, and documentation references. No security concerns present.
Pure documentation skill with no executable code. Contains only markdown reference files providing scientific writing guidance. No file access, network operations, or code execution capabilities detected.
9
Files scanned
4,242
Lines analyzed
0
Review items
0
False positives ignored
No confirmed security findings were recorded for this completed audit.