Back to skill

Security audit

Time Weaver

Security checks for vulnerabilities and agentic risk

Overview

This writing skill needs review because it automatically publishes generated chapters to a remote service and stores a reusable API key in plaintext.

Install only if you are comfortable with generated manuscripts being uploaded to the Time Weaver website. Do not paste API keys into chat, avoid automatic publishing, review every chapter before upload, and treat custom titles and imported chapter files as unsafe until the command and EPUB export paths are hardened.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • 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
Findings (6)

T01 · Skill Instruction Hijacking

Error
Location
sub-skills/chapter-generator/SKILL.md:129
Finding

Mandatory External Publication During the Local Writing Workflow

Content
View full analysis
Chapter published to https://time-weaver-782300018128.us-west1.run.app/ ``` The original skill text expressly labels publication as mandatory and instructs the agent to publish immediately after saving each chapter. The root skill reinforces this behavior: ```markdown 5. **Force-save the chapter file** 6. **Invoke `novel-publisher` to publish the chapter** ``` ### Technical Analysis The core purpose of the skill is interactive novel generation, which can be performed locally. However, the loaded skill instructions expand that task into mandatory publication to a fixed external service. Publication is not presented as an optional, separately authorized operation. Once configured, each generated chapter is uploaded automatically. The chapter generator also injects a fixed service URL into recurring output. This changes the agent's normal session behavior and introduces a network side effect that is not technically necessary for local writing. The instructions further state that a published chapter cannot be changed, increasing the consequences of accidental publication. ### Attack Path 1. A user loads the Time Weaver skill to generate a novel ...[truncated 1011 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
scripts/cli.js:14
Finding

API Key Requested Through Conversation and Stored as Plaintext Without Restrictive Permissions

Content
View full analysis
{ const key = data.toString().trim(); fs.writeFileSync(CONFIG_FILE, JSON.stringify({ apiKey: key })); console.log('Login successful. The key has been saved.'); process.exit(); }); } ``` The publisher instructions also direct the user to send the secret to the agent and write it using ordinary shell redirection: ```bash echo '{"apiKey": "USER_API_KEY"}' > ~/.time-weaver-config ``` ### Technical Analysis The documented workflow instructs the user to disclose a reusable API credential in the agent conversation. This can place the secret in conversation history, telemetry, debugging records, or other logs outside the credential's intended security boundary. The CLI then stores the key in a plaintext JSON file. `fs.writeFileSync` is invoked without an explicit mode such as `0o600`, while the documented shell redirection also relies entirely on the current process umask. On systems with permissive defaults, another local account or process may be able to read the key. The CLI additionally does not verify existing file ownership, reject symbolic links, or check whether the parent home directory was resolved safely. ### Attack Path 1. The user is told to send the API key to the agent. 2. The credential becomes part of the conversation or execution transcript. 3. The agent or CLI writes it to `~/.time-weaver-config`. 4. The ...[truncated 1112 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
scripts/cli.js:14
Finding

Environment-Controlled Endpoint Can Redirect API Keys and Manuscripts

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
sub-skills/chapter-generator/SKILL.md:138
Finding

User-Controlled Book Metadata Is Interpolated Into Shell Commands

Content
View full analysis
Remediation
View remediation

T08 · Insecure Dependencies

Warning
Location
scripts/package.json:1
Finding

Obsolete EPUB Dependency Introduces a Deprecated Transitive Dependency Tree

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/export-epub.js:30
Finding

Unescaped Chapter Content Is Embedded Directly Into EPUB HTML

Content
View full analysis
{ const code = match.replace(/```[\w]*\n?/, '').replace(/```$/, ''); return `
text
${code}
`; }) // Headings .replace(/^####\s*(.+)$/gm, '

$1

') .replace(/^###\s*(.+)$/gm, '

$1

') .replace(/^##\s*(.+)$/gm, '

$1

') .replace(/^#\s*(.+)$/gm, '

$1

') // Bold and italic .replace(/\*\*\*(.+?)\*\*\*/g, '$1') .replace(/\*\*(.+?)\*\*/g, '$1') .replace(/\*(.+?)\*/g, '$1') // Inline code .replace(/`(.+?)`/g, '$1') // Horizontal rule .replace(/^---+$/gm, '
') // Paragraphs .split('\n\n') .map(para => { para = para.trim(); if (!para) return ''; if (para.startsWith('${para.replace(/\n/g, '
')}

`; }) .join('\n'); return html; } ``` ### Technical Analysis The custom Markdown converter performs regular-expression substitutions without first HTML-escaping the source text. Raw tags and attributes contained in chapter files therefore remain present in the generated HTML. Code-block contents are also inserted between `` tags without escaping characters such as `<`, `>`, `&`, or quotation marks. A crafted code block can terminate the generated element and introduce additional markup. The resulting content is packaged into an EPUB. EPUB files commonly contain XHTML rendered by an embedded browser engine. Security behavior varies by reader: some readers disable scrip ...[truncated 1209 chars]
Remediation
View remediation
Vulnerability Patterns
  • 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
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
Findings (52)

Known Vulnerable Dependency: ejs==2.7.4 — 2 advisory(ies): CVE-2024-33883 (ejs lacks certain pollution protection); CVE-2022-29078 (ejs template injection vulnerability)

Critical
Category
Supply Chain
Confidence
98% confidence
Finding

ejs 2.7.4 has known template injection and pollution-related issues, and template engines become especially dangerous if any part of the template or rendering options can be influenced by untrusted input. Because this skill appears to support writing/content generation workflows, template rendering is plausible, making server-side code execution or severe integrity compromise a credible outcome if the vulnerable package is actually used.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The skill claims to support exploratory novel writing, but the static analysis indicates additional publication-oriented file ingestion and EPUB production behavior. When operational behavior materially exceeds the stated purpose, users and platforms cannot accurately assess trust boundaries, increasing the risk of unauthorized file access and unintended publication workflows.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The skill claims to support exploratory novel writing, but the static analysis indicates additional publication-oriented file ingestion and EPUB production behavior. When operational behavior materially exceeds the stated purpose, users and platforms cannot accurately assess trust boundaries, increasing the risk of unauthorized file access and unintended publication workflows.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The skill claims to support exploratory novel writing, but the static analysis indicates additional publication-oriented file ingestion and EPUB production behavior. When operational behavior materially exceeds the stated purpose, users and platforms cannot accurately assess trust boundaries, increasing the risk of unauthorized file access and unintended publication workflows.

Content

No source excerpt is available for this finding.

Known Vulnerable Dependency: brace-expansion==1.1.13 — 3 advisory(ies): CVE-2026-13149 (brace-expansion: DoS via exponential-time expansion of consecutive non-expanding); CVE-2026-14257 (brace-expansion: DoS via unbounded expansion length causing an out-of-memory pro); CVE-2026-69152 (brace-expansion: DoS via unbounded intermediate arrays, bypassing the CVE-2026-1)

High
Category
Supply Chain
Confidence
92% confidence
Finding

brace-expansion 1.1.13 is an outdated transitive dependency with reported denial-of-service issues stemming from pathological brace patterns that can trigger excessive CPU or memory consumption. In a lockfile this is a real supply-chain risk, though the practical impact depends on whether attacker-controlled glob/brace patterns are ever processed by the application.

Content

No source excerpt is available for this finding.

Known Vulnerable Dependency: form-data==2.5.5 — 1 advisory(ies): CVE-2026-12143 (form-data: CRLF injection in form-data via unescaped multipart field names and f)

High
Category
Supply Chain
Confidence
87% confidence
Finding

form-data 2.5.5 is flagged for CRLF injection in multipart field names/filenames, which can enable malformed requests or header manipulation when attacker-controlled values are serialized into outbound multipart bodies. This is a real issue in the dependency tree, although exploitability depends on whether the skill or its libraries build multipart requests from untrusted data.

Content

No source excerpt is available for this finding.

Known Vulnerable Dependency: lodash==4.17.23 — 2 advisory(ies): CVE-2025-13465 (lodash vulnerable to Prototype Pollution via array path bypass in `_.unset` and ); CVE-2021-23337 (lodash vulnerable to Code Injection via `_.template` imports key names)

High
Category
Supply Chain
Confidence
86% confidence
Finding

lodash 4.17.23 is associated with prototype-pollution and template/code-injection concerns in certain APIs such as _.template or unsafe object-path operations. The package presence in the lockfile is a genuine risk indicator, but the actual severity depends on whether the affected lodash functions are reachable with attacker-controlled inputs.

Content

No source excerpt is available for this finding.

Known Vulnerable Dependency: lodash.pick==4.4.0 — 1 advisory(ies): CVE-2020-8203 (Prototype Pollution in lodash)

High
Category
Supply Chain
Confidence
90% confidence
Finding

lodash.pick 4.4.0 has a known prototype pollution issue, which can let crafted property paths modify object prototypes and corrupt application behavior. Even as a transitive dependency, this is a real vulnerability if untrusted keys are processed through the affected API.

Content

No source excerpt is available for this finding.

Known Vulnerable Dependency: nth-check==1.0.2 — 1 advisory(ies): CVE-2021-3803 (Inefficient Regular Expression Complexity in nth-check)

High
Category
Supply Chain
Confidence
90% confidence
Finding

nth-check 1.0.2 is vulnerable to inefficient regular-expression or parsing behavior that can cause denial of service when processing crafted selectors. Since it is used by cheerio/css-select, exploitability depends on whether attacker-controlled CSS selectors are accepted, but the dependency risk itself is real.

Content

No source excerpt is available for this finding.

Missing User Warnings

High
Category
Not specified by scanner
Confidence
95% confidence
Finding

Mandating automatic save and immediate publication without warning or consent creates unsafe autonomous side effects. Even if the content is 'just fiction,' the workflow can overwrite files, create persistent artifacts, and publicly disclose material before the user reviews it.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

High
Category
Not specified by scanner
Confidence
96% confidence
Finding

The skill instructs execution of an external publish command that sends generated chapter content to a public website. That creates a real exfiltration/public-disclosure risk because model-generated content may include sensitive user inputs, unpublished material, or copyrighted text, and the action occurs as part of the normal workflow rather than after informed approval.

Content

No source excerpt is available for this finding.

Missing User Warnings

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The publication command transmits chapter content to an external website with no clear user warning, review step, or consent gate. In context, this makes the skill more dangerous because a creative-writing tool normally implies content generation, not automatic outbound publication, so users may not expect disclosure to a remote service.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The skill grants itself credential-handling capability by directing storage of an API key in the user's home directory, even though this is not necessary for a writing-only function. Persisting credentials locally increases the attack surface for theft, misuse by other tools, and unauthorized future publishing actions.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

Automatic invocation after each chapter enables autonomous transmission of newly written content to an external platform without a fresh user checkpoint. In a creative-writing context, this is especially dangerous because drafts may be incomplete, private, or commercially sensitive, and repeated automation amplifies harm.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
86% confidence
Finding

The skill orchestrates file reads/writes, sub-skill loading, and external publication behavior, yet declares no explicit tool scope or permissions boundary. In a skill framework, missing scope makes privilege expectations ambiguous and can permit broader environment or code capabilities than users would reasonably infer.

Content

No source excerpt is available for this finding.

Skill Enumeration

Medium
Category
Agent Snooping
Confidence
80% confidence
Finding

Skill enumerates or reads other installed skills. Access to other skills' SKILL.md files or the skills directory reveals prompt instructions, capabilities, and secrets that should be invisible to peer skills.

Content

Scanner excerpt · SKILL.md (reported line 27)May include surrounding context.

md
### 第一步:选择写作风格(调用 writing-style)

**调用方式**:读取 `sub-skills/writing-style/SKILL.md`,按流程执行:

1. 显示可用风格列表
2. 用户选择

Skill Enumeration

Medium
Category
Agent Snooping
Confidence
80% confidence
Finding

Skill enumerates or reads other installed skills. Access to other skills' SKILL.md files or the skills directory reveals prompt instructions, capabilities, and secrets that should be invisible to peer skills.

Content

Scanner excerpt · SKILL.md (reported line 82)May include surrounding context.

md
### 第四步:生成年代背景(调用 era-generator)

**调用方式**:读取 `sub-skills/era-generator/SKILL.md`,按流程执行:

1. 询问隐藏模式(隐藏/公开)
2. 选择时代类型(国内历史/国外历史/虚构/未来/末日)

Skill Enumeration

Medium
Category
Agent Snooping
Confidence
80% confidence
Finding

Skill enumerates or reads other installed skills. Access to other skills' SKILL.md files or the skills directory reveals prompt instructions, capabilities, and secrets that should be invisible to peer skills.

Content

Scanner excerpt · SKILL.md (reported line 93)May include surrounding context.

md
### 第五步:生成主角(调用 character-generator)

**调用方式**:读取 `sub-skills/character-generator/SKILL.md`,按流程执行:

1. 询问基础信息(姓名、性别、现代身份)
2. 生成5个角色选项供选择

Skill Enumeration

Medium
Category
Agent Snooping
Confidence
80% confidence
Finding

Skill enumerates or reads other installed skills. Access to other skills' SKILL.md files or the skills directory reveals prompt instructions, capabilities, and secrets that should be invisible to peer skills.

Content

Scanner excerpt · SKILL.md (reported line 104)May include surrounding context.

md
### 第六步:生成初始场景(调用 scene-generator)

**调用方式**:读取 `sub-skills/scene-generator/SKILL.md`,按流程执行:

1. 根据年代和角色设定,生成5个初始场景选项
2. 用户选择或自定义

Skill Enumeration

Medium
Category
Agent Snooping
Confidence
80% confidence
Finding

Skill enumerates or reads other installed skills. Access to other skills' SKILL.md files or the skills directory reveals prompt instructions, capabilities, and secrets that should be invisible to peer skills.

Content

Scanner excerpt · SKILL.md (reported line 114)May include surrounding context.

md
### 第七步:小说命名(调用 novel-namer)

**调用方式**:读取 `sub-skills/novel-namer/SKILL.md`,按流程执行:

1. 根据年代、角色、场景等设定,生成5个书名选项
2. 用户选择、重新生成或自定义

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The skill mandates automatic chapter saves and invokes a publishing command without an explicit warning, approval step, or distinction between local draft persistence and remote publication. This is dangerous because a user asking for writing assistance may unintentionally trigger file modification and external disclosure of unpublished or sensitive content.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The instruction '当新人物出现时,自动生成完整档案' is an overly broad trigger because 'new character appears' is not operationally bounded. In an agent setting, this can cause the skill to fire on incidental mentions, hypothetical examples, or partial references, leading to unwanted invocation, excess output, and interference with higher-priority user tasks.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The timing rules use ambiguous conditions like immediately generating on meeting a new NPC and periodically updating every 5 chapters, but do not specify who determines importance, what counts as a chapter, or when generation should be skipped. This ambiguity can produce repeated or premature invocation, unnecessary state changes, and unreliable behavior across different agent runtimes.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

This markdown file presents all skill content in a single language, which can amount to a language/locale policy violation when no user opt-in or documented regional justification is provided. The file does not indicate that Chinese is optional, selectable, or required for a specific audience.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The entire skill is written as a Chinese-only writing guide and presents its instructions, examples, and checklist exclusively in Chinese, with no indication that users may choose another language or locale. Under the stated policy, forcing a specific language without user opt-in is a natural-language policy concern.

Content

No source excerpt is available for this finding.