Back to skill

Security audit

Consistency Checker

Security checks for vulnerabilities and agentic risk

Overview

The documented checker is mostly local, but the package also includes an under-disclosed LLM script that can upload manuscript excerpts and character data to an external service.

Install only if you are comfortable reviewing the scripts first and avoiding or modifying the LLM variant unless you intentionally want to send manuscript excerpts and character information to DashScope. For private or unpublished writing, use the documented local checker only, pin dependencies in an isolated environment, and require an explicit opt-in before any remote LLM call.

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 (2)

other

Warning
Location
scripts/check_consistency_llm.py:90
Finding

Undisclosed Transmission of Manuscript and Character Data to an External LLM Service

Content
View full analysis
Remediation
View remediation

T08 · Insecure Dependencies

Note
Location
scripts/requirements.txt:1
Finding

Unpinned and Incomplete Third-Party Dependency Specification

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Taint TrackingDirect Taint Flow, Variable-Mediated Taint Flow, Credential Exfiltration Chain
Findings (13)

Tainted flow: 'headers' from os.environ.get (line 127, credential/environment) → requests.post (network output)

Critical
Category
Data Flow
Confidence
90% confidence
Finding

Credentials or environment variables flow to a network sink. This is a high-confidence indicator of credential exfiltration.

Content

Scanner excerpt · scripts/check_consistency_llm.py (reported line 151)May include surrounding context.

python
transient=True,
    ) as progress:
        progress.add_task("checking", total=None)
        response = requests.post(url, headers=headers, json=payload, timeout=90)

    response.raise_for_status()
    data = response.json()

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
70% confidence
Finding

Without declared permissions the skill's intent is opaque and cannot be validated.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The manifest description and title are written to present the skill as a Chinese-language checker, and the rest of the document consistently assumes Chinese inputs and outputs. There is no explicit user opt-in, language choice, or documented justification that this skill is intended only for Chinese-language novels, which can violate language/locale policy requirements.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The module docstring is entirely in Chinese and presents the tool as a Chinese-language checker, with no indication that language choice is optional. Under the policy, forcing a specific language or locale without user opt-in is a natural-language policy violation unless the constraint is explicitly documented and justified.

Content

No source excerpt is available for this finding.

Ssd 4

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

Untrusted novel content is embedded directly into the prompt sent to a high-authority LLM, so adversarial text inside chapters can influence the model's behavior or output format. While this does not directly execute code, it can cause prompt injection effects such as ignoring instructions, producing malformed JSON, or returning manipulated findings that degrade integrity of the analysis.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The skill is presented as a local consistency checker, but it actually sends chapter text and character profiles to an external LLM service. That creates a data exposure risk because sensitive unpublished manuscript content or personal notes may leave the local environment without the user's explicit awareness.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
95% confidence
Finding

The code performs external network transmission of manuscript and character data to a third-party service. In this skill context, that is security-relevant because the transmitted content may be sensitive, unpublished, or proprietary, and the skill metadata does not make that behavior obvious.

Content

Scanner excerpt · scripts/check_consistency_llm.py (reported line 151)May include surrounding context.

python
transient=True,
    ) as progress:
        progress.add_task("checking", total=None)
        response = requests.post(url, headers=headers, json=payload, timeout=90)

    response.raise_for_status()
    data = response.json()

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

The script transmits chapter content and character data to an external LLM API without a clear runtime warning or consent gate. Users may unintentionally upload confidential drafts, proprietary worldbuilding, or sensitive notes, which is a meaningful privacy and confidentiality risk.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
83% confidence
Finding

The file presents its description, prompts, console messages, and model instructions exclusively in Chinese. This can violate language/locale policy when a skill forces one language without opt-in or documented justification.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Low
Category
Not specified by scanner
Confidence
76% confidence
Finding

A novel consistency checker's stated purpose is text consistency analysis, but the implementation depends on reading a credential from the environment to access an external provider. That credential access is not obviously implied by the manifest description and introduces an additional capability beyond the user-facing purpose.

Content

No source excerpt is available for this finding.

Unpinned Dependencies

Low
Category
Supply Chain
Confidence
96% confidence
Finding

The dependency rich is unpinned, so builds may resolve to different versions over time, creating supply-chain instability and the possibility of pulling in a compromised or breaking release. This is a real security weakness, though low severity by itself, because reproducibility and controlled patching are lost.

Content

Scanner excerpt · scripts/requirements.txt (reported line 1)May include surrounding context.

text
rich
PyYAML

Unpinned Dependencies

Low
Category
Supply Chain
Confidence
99% confidence
Finding

PyYAML is unpinned, which is more concerning because multiple known advisories exist for some releases of this package. Without a pinned version, deployments may install an affected release or drift between safe and unsafe versions, increasing supply-chain and vulnerability exposure.

Content

Scanner excerpt · scripts/requirements.txt (reported line 2)May include surrounding context.

text
rich
PyYAML

Unverifiable Dependency: PyYAML has 8 known advisory(ies) (CVE-2019-20477 (Deserialization of Untrusted Data in PyYAML); CVE-2020-1747 (Improper Input Validation in PyYAML); CVE-2020-14343 (Improper Input Validation in PyYAML) +5 more), but the manifest does not pin a version, so it is unknown whether the installed release is affected

Low
Category
Supply Chain
Confidence
98% confidence
Finding

PyYAML has multiple known advisories affecting some versions, and because no version is pinned, it is impossible to verify whether the installed dependency is patched. In a content-processing skill like a consistency checker, YAML may plausibly be used for configuration or data ingestion, so leaving the version unconstrained increases the chance of deploying a vulnerable parser.

Content

No source excerpt is available for this finding.