Back to skill

Security audit

Scisurvey

Security checks for vulnerabilities and agentic risk

Overview

The skill is a coherent academic-survey generator, but its document export path can run local shell commands with unsanitized user-derived filenames.

Install only if you are comfortable with a skill that may create local review files and, for LaTeX/DOCX/PDF export, invoke local document-conversion tools. Use simple safe filenames, prefer Markdown output, and require explicit approval before any shell command or file conversion runs.

Vulnerability Patterns
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
Findings (1)

T09 · Insecure Skill Coding Practices

Warning
Location
SKILL.md:1380
Finding

Unsanitized User-Controlled Topic Interpolation in Shell Commands

Content
View full analysis
/dev/null && COMPILER=xelatex || which pdflatex 2>/dev/null && COMPILER=pdflatex # Chinese content requires xelatex $COMPILER -review.tex bibtex -review $COMPILER -review.tex $COMPILER -review.tex # Run twice to resolve citations ``` ```bash # Basic DOCX conversion pandoc -review.md \ --from markdown \ --to docx \ --reference-doc ~/.claude/skills/sciverse-review/templates/reference.docx \ --bibliography -review.bib \ --csl ~/.claude/skills/sciverse-review/csl/.csl \ -o <topic>-review.docx ``` ```bash pandoc <topic>-review.md \ --from markdown \ --to pdf \ --pdf-engine=xelatex \ --bibliography <topic>-review.bib \ --csl ~/.claude/skills/sciverse-review/csl/<style>.csl \ -V CJKmainfont="PingFang SC" \ -V geometry:margin=2.5cm \ -o <topic>-review.pdf ``` ### Technical Analysis The skill derives `<topic>` from the user's requested review subject and uses it to construct filenames passed to shell commands. The instructions do not require the topic to be normalized, validated, escaped, or shell-quoted before substitution. If an agent replaces `<topic>` with the original user-controlled text and executes the resulting command through a shell, metacharacters such as semicolons, command substitutions, redirection operators, or shell control operators may be interpreted as command syntax rather than as part of a filename. The same unsafe pattern affects the LaTeX, BibTeX, DOCX, and PDF conversion paths. The `<style>` placeholder is also unquoted and should be constrained to the documented allowlist, although the confirmed primary input path is the user-supplied topic. This issue does not itself provide privilege escalation. Su ...[truncated 1722 chars]
Remediation
View remediation
apa-7th-edition.csl ieee -> ieee.csl vancouver -> vancouver.csl chicago -> chicago-author-date.csl mla -> modern-language-association-9th-edition.csl gbt -> gb-t-7714-2015-numeric.csl ``` 6. **Separate display titles from filesystem names** Preserve the original topic only inside document content. Never reuse the display title as a command argument or filesystem path without normalization. 7. **Apply least privilege** Run document-conversion tools in a restricted working directory under an unprivileged account, with only the filesystem and network access required for report generation. ...[truncated 5 chars]
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (6)

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The activation examples use broad phrases like asking for a survey or literature review, which can cause the skill to trigger on generic user requests without the user understanding that a powerful automation workflow may start. Because this skill includes extensive retrieval, file generation, and potential local tool invocation, overbroad triggering raises the chance of surprising or unauthorized actions.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The metadata declares no required binaries, but later instructions depend on local executables including xelatex, pdflatex, bibtex, and pandoc. This mismatch hides the skill's real execution requirements, undermines informed consent and security review, and can cause the agent to attempt unexpected local tool usage in environments where such execution was not anticipated.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

The writing rules state '语言:客观学术中文' and later reinforce '中文综述为主:正文用中文', which imposes a specific language/locale by default. The file does not clearly offer users a choice of output language or require consent before enforcing Chinese output.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The skill describes writing Markdown, LaTeX, DOCX, PDF, and BibTeX outputs and invoking conversion/compilation tools, but the user-facing description does not clearly warn that local files may be created and external binaries may be run. This lack of disclosure weakens user consent and can lead to unexpected filesystem changes or execution of local tooling.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill instructs the agent to write files and execute local shell commands invoking document-conversion and compilation toolchains such as xelatex, pdflatex, bibtex, and pandoc. For a literature-survey skill, this expands capability from content generation into local code/tool execution, increasing risk of unintended command execution, file overwrites, data exposure through auxiliary tools, or abuse of complex parsers/compilers.

Content

No source excerpt is available for this finding.

Sudo/Root Execution

Medium
Category
Privilege Escalation
Confidence
70% confidence
Finding

Commands invoke sudo or root privileges. Verify this elevated access is necessary and justified.

Content

Scanner excerpt · SKILL.md (reported line 1423)May include surrounding context.

gbt → gb-t-7714-2015-numeric.csl(或 -author-date)

text

> 若 Pandoc 不可用:告知用户 `brew install pandoc`(macOS)或 `sudo apt install pandoc`(Linux),并提供等效命令;同时交付 `.md` 源文件作为备选。

---

Static analysis

No suspicious patterns detected.