Back to skill

Security audit

hit-content-writer

Security checks for vulnerabilities and agentic risk

Overview

This is a coherent content rewriting skill, but it can persist and retrieve user content, send text to an external embedding service, and install an unpinned tool without enough user control or disclosure.

Review this before installing if you may submit confidential drafts, personal data, client content, or copyrighted source material. Use explicit skill-name commands, avoid generic storage/search prompts, install a reviewed pinned version of `jl-vector-store` in a least-privilege environment, use a scoped embedding API key, and confirm where ChromaDB data is stored and how to delete it.

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:36
Finding

Unpinned Third-Party Tool Installation

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, line 36
Vulnerability Type: Unpinned third-party dependency installation
Risk Level: Medium

Vulnerable Code

bash
uv tool install jl-vector-store

The same unsafe installation instruction is duplicated in README.md, line 21:

bash
uv tool install jl-vector-store

Technical Analysis

The project instructs users to install jl-vector-store without specifying an exact version, package hash, signature, or authoritative package repository. Consequently, the package resolved at installation time may differ from the package that was originally reviewed or tested.

Package installation can execute package-controlled build or installation logic. The installed command is also subsequently trusted to process content and interact with an embedding service. If the package, its distribution account, or one of its transitive dependencies is compromised, attacker-controlled code could execute with the privileges of the user performing the installation.

This is a supply-chain weakness rather than evidence that the current jl-vector-store package is malicious.

Attack Path

  1. An attacker compromises the package publisher, package repository entry, or a transitive dependency used by jl-vector-store.
  2. The attacker publishes a malicious release that remains compatible with the unversioned package name.
  3. A user follows the documented uv tool install jl-vector-store instruction.
  4. The package manager resolves and installs the attacker-controlled release because no trusted version or integrity constraint is specified.
  5. Malicious installation or runtime logic executes under the installing user's account.
  6. The malicious package may access data and credentials available to that account, including content sent for storage and embedding-service environment variables.

Impact Assessment

Successful exploitation could provide arbitrary code execution with the privileges of the u ...[truncated 571 chars]

Remediation
View remediation

Remediation Suggestions

  1. Pin jl-vector-store to a specifically reviewed version:
    bash
    uv tool install "jl-vector-store==<audited-version>"
    
  2. Generate and verify cryptographic hashes for the package and all transitive dependencies where the package-management workflow supports this.
  3. Explicitly identify and configure the trusted package index rather than relying on ambient package-manager configuration.
  4. Document the package's authoritative source repository, publisher identity, and release-verification procedure.
  5. Audit the pinned package and its dependency tree before recommending installation.
  6. Install and run the tool in an isolated, least-privilege environment with access only to required files and network destinations.
  7. Avoid exposing unrelated credentials to the tool process. Provide the embedding API key only when needed and use a narrowly scoped, revocable credential.
  8. Add a controlled dependency-update process that reviews and tests new versions before changing the pin.
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (19)

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The README describes storing user content in a vector database and configuring an external embedding API, but it does not clearly warn that user-provided text may be persisted and transmitted to a third-party service. This creates a meaningful privacy and data-governance risk, especially if users submit proprietary, personal, or regulated content assuming local-only processing.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
88% confidence
Finding

The skill is configured to use an external endpoint for embeddings, which means content submitted for storage/query workflows may be transmitted outside the local environment. External transmission is not inherently malicious, but without clear consent boundaries and data classification guidance it can expose sensitive text to third-party infrastructure.

Content

Scanner excerpt · README.md (reported line 26)May include surrounding context.

  1. 环境变量配置(用于向量化API):
bash
export EMBEDDING_BASE_URL="https://api.siliconflow.cn/v1/embeddings"
export EMBEDDING_API_KEY="your-api-token"
export EMBEDDING_MODEL="BAAI/bge-large-zh-v1.5"

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The extraction trigger phrases include very generic terms like '提取文案' and '获取文案', which can plausibly appear in ordinary conversation and cause unintended invocation of the skill. In an agent environment, broad triggers increase the risk of prompt/command collision, causing the system to process links or content the user did not intend to send through this skill path.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The analysis triggers are common-language phrases like '分析内容' and '智能分析', making accidental activation likely during normal user requests. That can route arbitrary text or file paths into this skill unexpectedly, expanding the attack surface for unintended data handling.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The storage triggers are broad phrases such as '保存内容' and '存储内容', which may overlap with routine requests and cause unintentional persistence of user data. Because this operation stores content and may forward it for embedding, accidental activation has higher consequences than a read-only feature.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

Query phrases like '查询内容' and '搜索内容' are too generic and can match ordinary search requests intended for other tools or the base agent. This can lead to unintended retrieval from the vector database and disclosure of previously stored content to the wrong conversational context.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

Rewrite triggers such as '仿写内容' and '智能仿写' are broad enough to collide with normal writing-help requests. In agent orchestration, this may unintentionally invoke transformation logic on sensitive user text or cause outputs to be generated under assumptions the user did not explicitly approve.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The trigger phrases are very broad and map to common requests like extracting, analyzing, saving, searching, or rewriting content. This creates a skill-squatting risk where normal user intent could invoke this skill unexpectedly, causing unrequested handling of external links, local files, or storage operations. The context increases risk because the skill can call dependent tools and perform side-effecting actions such as downloading content and storing it in a vector database.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The storage feature sends user-provided content into a vector database but the skill does not present a clear warning at the point of use about persistence, retention, or sensitivity of stored material. Users may provide private, confidential, or copyrighted content without realizing it will be embedded and retained for later retrieval, creating privacy and data-governance risks. The skill context makes this more dangerous because storage is presented as a simple standalone action and examples encourage direct ingestion of arbitrary text or files.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The entire skill design is written around Chinese-language processing, including Chinese tokenization with jieba and Chinese-specific sentence splitting, but the document does not state that this is a China/Chinese-only skill or offer any user opt-in for language handling. Because the policy requires avoiding forced language or locale constraints unless documented and justified, this reads as an implicit Chinese-only constraint.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The phrase indicating the analyzer is 'automatically called when the user requests content analysis' is underspecified and may permit over-broad invocation based on loosely interpreted user intent. In an agent skill, ambiguous auto-triggering can cause unintended processing of user content or context, increasing privacy risk and making downstream actions occur without clear, explicit consent boundaries.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The guide gives users explicit rewriting workflows, similarity targets, and prompts for producing 'similar' or 'viral' content, but it does not include meaningful safeguards about copyright, plagiarism, attribution, or avoiding substantial similarity to source material. In the context of a content-rewriting skill, this omission increases the likelihood that users will generate infringing derivative works or deceptive near-copies, especially because the document normalizes structure imitation and reuse of successful patterns.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The document promotes imitation-based rewriting and even specifies target similarity ranges, but it does not warn users about copyright infringement, plagiarism, or derivative-work risks. In this context, the absence of guardrails can normalize copying protected expression and increase the chance that the skill is used to generate infringing content at scale.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The workflow says the skill queries reference content and later 'learns' from user content and feedback, but provides no privacy notice, consent model, retention limits, or restrictions on using submitted text for future training. This can lead to unintended collection, storage, or reuse of sensitive user data, especially if users submit drafts, proprietary materials, or personal stories for rewriting.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

This markdown template is entirely written in Chinese and labels the report format accordingly, which implicitly constrains the skill's output language. The file does not provide any user opt-in, alternative locale, or justification that the skill is intended only for a Chinese-language or region-specific context.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The entire template is written as a fixed Chinese output format, including headings, field labels, and status values, with no indication that users can select another language or opt in to Chinese. This is a natural-language locale policy concern because the file appears to prescribe a single language for all generated results without documented justification.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
90% confidence
Finding

This markdown file uses a single forced locale throughout and does not mention that the language is configurable or intended only for a Chinese-speaking audience. Under the policy, forcing a specific language without user opt-in is a natural-language policy concern.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
75% confidence
Finding

All headings, instructions, and examples in the skill reference are presented only in Chinese, which can amount to a language/locale restriction without user opt-in. The file does not indicate that the skill is region-specific or provide an alternative language option.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
83% confidence
Finding

The entire skill file presents instructions and interaction formats only in Chinese, with no indication that users may choose another language or that the skill is intentionally limited to a Chinese-speaking audience. This can constitute a language/locale policy issue under the requirement to avoid forcing a specific language without user opt-in.

Content

No source excerpt is available for this finding.