Back to skill

Security audit

Literature Review

Security checks for vulnerabilities and agentic risk

Overview

This literature-review skill uses expected academic search, citation-checking, and PDF-generation tools, with some disclosure and dependency hygiene issues but no evidence of malicious behavior.

Install only if you are comfortable sending literature search queries and citation identifiers to public academic metadata services. Use a virtual environment, pin dependencies where possible, avoid running PDF generation on untrusted markdown or bibliography files, and treat the AI-schematic requirement as optional unless you have separately installed and trust that other skill/provider.

Vulnerability Patterns
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • 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)

T08 · Insecure Dependencies

Warning
Location
SKILL.md:552
Finding

Unpinned Third-Party Dependency Installation

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 552-555
Vulnerability Type: Unpinned third-party dependency installation
Risk Level: Medium

Vulnerable Code:

markdown
### Required Python Packages
```bash
pip install requests  # For citation verification
text

### Technical Analysis

The Skill instructs users to install `requests` directly from the package index without specifying an audited version, cryptographic hashes, a lockfile, or an explicitly trusted index. The installed artifact and its transitive dependencies can therefore change independently of the reviewed Skill.

Package installation may execute package build logic with the invoking user's permissions. If the resolved package, one of its dependencies, or the configured package index is compromised, attacker-controlled code could execute during installation or later when the package is imported.

### Attack Path

1. An attacker compromises a future release in the resolved dependency chain, interferes with the configured package index, or otherwise causes `pip` to resolve an attacker-controlled distribution.
2. A user follows the documented `pip install requests` instruction.
3. `pip` downloads and installs the unverified package and its dependencies.
4. Malicious build, installation, or import-time code executes under the user's account.
5. The code can access resources available to that account, including project files and environment variables.

### Impact Assessment

Successful exploitation could provide arbitrary code execution with the privileges of the user running `pip` or the scripts. The scope may include reading or modifying accessible files, accessing environment variables such as configured academic API keys, making network requests, and altering the active Python environment. This instruction does not itself request elevated privileges, so the achievable privileges are bounded by the invoking account and environmen
...[truncated 2 chars]
Remediation
View remediation

Remediation Suggestions

  1. Pin reviewed direct and transitive dependency versions in a lockfile or requirements file.
  2. Record cryptographic hashes and require verification:
    bash
    python -m pip install --require-hashes -r requirements.txt
    
  3. Configure an explicitly trusted package index rather than relying on ambient pip configuration.
  4. Install dependencies inside an isolated virtual environment with minimal filesystem and credential access.
  5. Review and test dependency updates before changing pinned versions.
  6. Add automated dependency vulnerability and integrity scanning to the release process.
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Behavioral ASTexec() Call, eval() Call, Dynamic Import
  • Taint TrackingDirect Taint Flow, Variable-Mediated Taint Flow, Credential Exfiltration Chain
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (18)

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

Another materially similar mismatch is present here: the skill is represented as comprehensive and reproducible, but the described implementation does not support the full advertised workflow. In security terms, deceptive or overstated capability claims are risky because autonomous agents may chain this skill into higher-stakes pipelines and rely on nonexistent safeguards such as citation verification or reproducible search documentation.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

Another materially similar mismatch is present here: the skill is represented as comprehensive and reproducible, but the described implementation does not support the full advertised workflow. In security terms, deceptive or overstated capability claims are risky because autonomous agents may chain this skill into higher-stakes pipelines and rely on nonexistent safeguards such as citation verification or reproducible search documentation.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

Another materially similar mismatch is present here: the skill is represented as comprehensive and reproducible, but the described implementation does not support the full advertised workflow. In security terms, deceptive or overstated capability claims are risky because autonomous agents may chain this skill into higher-stakes pipelines and rely on nonexistent safeguards such as citation verification or reproducible search documentation.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

Another materially similar mismatch is present here: the skill is represented as comprehensive and reproducible, but the described implementation does not support the full advertised workflow. In security terms, deceptive or overstated capability claims are risky because autonomous agents may chain this skill into higher-stakes pipelines and rely on nonexistent safeguards such as citation verification or reproducible search documentation.

Content

No source excerpt is available for this finding.

Lp1

High
Category
MCP Least Privilege
Confidence
98% confidence
Finding

The script makes outbound network requests to multiple third-party services, but the skill metadata does not declare network capability/permission. Undeclared network access is dangerous in agent systems because user queries and identifiers are transmitted externally without an explicit permission boundary or operator awareness.

Content

No source excerpt is available for this finding.

Lp1

High
Category
MCP Least Privilege
Confidence
98% confidence
Finding

The script makes outbound network requests to multiple third-party services, but the skill metadata does not declare network capability/permission. Undeclared network access is dangerous in agent systems because user queries and identifiers are transmitted externally without an explicit permission boundary or operator awareness.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The guidance states 'Language (typically English, or specify multilingual)', which establishes a default language preference in natural-language instructions. Under the policy, forcing or biasing toward a specific language without explicit user choice can be a locale/language policy violation unless clearly justified.

Content

No source excerpt is available for this finding.

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
70% confidence
Finding

subprocess module calls execute external commands. Without careful input validation, this enables command injection.

Content

Scanner excerpt · scripts/generate_pdf.py (reported line 46)May include surrounding context.

python
# Check if pandoc is installed
    try:
        subprocess.run(['pandoc', '--version'], capture_output=True, check=True)
    except (subprocess.CalledProcessError, FileNotFoundError):
        print("Error: pandoc is not installed.")
        print("Install with: brew install pandoc (macOS) or apt-get install pandoc (Linux)")

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
86% confidence
Finding

This subprocess call executes pandoc on user-influenced inputs such as the markdown file, bibliography, citation style, and optional template. Although shell injection is mitigated by passing a list to subprocess.run, pandoc/xelatex may process untrusted content, local templates, and LaTeX constructs that can trigger file reads, network fetches, or other unsafe behavior depending on pandoc and TeX configuration. In a literature-review skill that converts externally sourced documents, this materially increases the attack surface.

Content

Scanner excerpt · scripts/generate_pdf.py (reported line 91)May include surrounding context.

python
try:
        print(f"Generating PDF: {output_pdf}")
        print(f"Command: {' '.join(cmd)}")
        result = subprocess.run(cmd, capture_output=True, text=True, check=True)
        print(f"✓ PDF generated successfully: {output_pdf}")
        return True
    except subprocess.CalledProcessError as e:

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
70% confidence
Finding

subprocess module calls execute external commands. Without careful input validation, this enables command injection.

Content

Scanner excerpt · scripts/generate_pdf.py (reported line 110)May include surrounding context.

python
missing = []
    for name, cmd in dependencies.items():
        try:
            subprocess.run(cmd.split(), capture_output=True, check=True)
            print(f"✓ {name} is installed")
        except (subprocess.CalledProcessError, FileNotFoundError):
            print(f"✗ {name} is NOT installed")

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · scripts/lit_search.py (reported line 10)May include surrounding context.

python
import time
import xml.etree.ElementTree as ET

S2_BASE_URL = "https://api.semanticscholar.org/graph/v1"
OA_BASE_URL = "https://api.openalex.org"
CR_BASE_URL = "https://api.crossref.org/works"
PM_BASE_URL = "https://eutils.ncbi.nlm.nih.gov/entrez/eutils"

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · scripts/lit_search.py (reported line 12)May include surrounding context.

python
S2_BASE_URL = "https://api.semanticscholar.org/graph/v1"
OA_BASE_URL = "https://api.openalex.org"
CR_BASE_URL = "https://api.crossref.org/works"
PM_BASE_URL = "https://eutils.ncbi.nlm.nih.gov/entrez/eutils"
EPM_BASE_URL = "https://www.ebi.ac.uk/europepmc/webservices/rest"

Tainted flow: 'fetch_params' from requests.get (line 183, network input) → requests.get (network output)

Medium
Category
Data Flow
Confidence
65% confidence
Finding

Data from a source is assigned to a variable that is later passed to a sink, creating a variable-mediated taint flow.

Content

Scanner excerpt · scripts/lit_search.py (reported line 189)May include surrounding context.

python
"retmode": "xml",
            "rettype": "abstract"
        }
        fetch_res = requests.get(fetch_url, params=fetch_params, timeout=REQUEST_TIMEOUT)
        if fetch_res.status_code != 200:
            return {"error": f"efetch HTTP {fetch_res.status_code}", "data": []}

External Transmission

Medium
Category
Data Exfiltration
Confidence
97% confidence
Finding

The script transmits extracted DOIs from a local markdown file to third-party services (doi.org and CrossRef) without any consent prompt, offline mode, or disclosure controls. In a literature-review skill this behavior is functional, but it still leaks document-derived identifiers and access patterns to external services, which may expose sensitive research topics, unpublished references, or confidential manuscript contents indirectly.

Content

Scanner excerpt · scripts/verify_citations.py (reported line 47)May include surrounding context.

python
def _get_crossref_metadata(self, doi: str) -> Dict:
        """Get metadata from CrossRef API."""
        try:
            url = f"https://api.crossref.org/works/{doi}"
            response = self.session.get(url, timeout=10)

            if response.status_code == 200:

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
87% confidence
Finding

The template presents 'English language' as a default inclusion criterion, which can bias reviews toward English-only evidence without requiring justification or explicit opt-in. In a literature-review skill, this may systematically exclude relevant non-English studies and weaken completeness, reproducibility, and fairness of evidence synthesis.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
89% confidence
Finding

Line L140 recommends limiting searches to 'English (or specify multilingual)', which sets English as the default language criterion in natural-language guidance. Because the document does not provide explicit user opt-in or explain a region-specific need for English-only searching, this is a language/locale policy concern.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
93% confidence
Finding

Line L225 includes 'AND English[Language]' in the example query, which operationalizes an English-only search by default. Since the file does not frame this as optional or user-selected, it creates a natural-language locale restriction that may conflict with policy expectations around language choice.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
81% confidence
Finding

The script saves a detailed report to a new file automatically using a path derived from the provided markdown filename. Although this is part of the tool's functionality, there is no confirmation prompt or explicit disclosure near the write operation itself, so users may not realize a local file will be created or overwritten.

Content

No source excerpt is available for this finding.