Security audit
Security checks for vulnerabilities and agentic risk
This writing skill is not malicious, but it needs review because it automatically uses external search on user ideas and personal context without clear consent or a no-search default.
Install only if you are comfortable with the skill using external web search as part of drafting. Avoid giving it confidential business plans, unpublished product ideas, personal stories, third-party details, secrets, or sensitive workplace information unless you explicitly want that context used. For private brainstorming, ask for a no-search/local-only workflow before sharing details.
The README advertises "Intelligent Search Integration" that automatically searches for materials, but does not warn users that their prompts or derived topic details may be sent to external tools or services. This creates a privacy and data-governance risk, especially when users provide sensitive business ideas, personal experiences, or unpublished content assumptions during brainstorming.
The documented workflow includes both an initial background search and a later precision search, yet gives no warning that external data access may occur as part of normal operation. Because the skill is designed to progressively extract richer user context over 7-10 rounds, these searches may be informed by increasingly sensitive details, amplifying the privacy impact of implicit external sharing.
The README ends with a very broad trigger phrase, "I want to expand an idea...", which can overlap with ordinary conversation and cause unintentional invocation in environments where skills are selected from natural language. Because this skill performs multi-turn guidance and may initiate search behavior, accidental activation can lead to unnecessary tool use, unwanted data disclosure, or user confusion.
The manifest markets the skill as simple writing assistance, but the body requires mandatory external web research and validation. This creates a capability disclosure gap: users and orchestrators may invoke the skill expecting local text expansion while it actually sends topic content to external services, increasing privacy and policy risk.
The trigger phrases are broad enough to match generic brainstorming, writing, and thinking-assistance requests, which can cause over-invocation of a skill that also performs repeated questioning and external search. Over-broad routing increases the chance that unrelated or sensitive user content is sent into this workflow without clear necessity.
Skill selects an external model or provider that may use a different account or billing plan than the operator expects. Undisclosed model switches can cause unexpected cost or quota consumption.
description: Expand scattered ideas into in-depth posts. Use this skill when users need to "expand ideas," "generate posts," "brainstorm content," "writing assistance," "enrich thoughts," "deepen thinking," or provide scattered ideas that need systematic expression.
version: 1.5.0
allowed-tools: AskUserQuestion, WebSearch, mcp__web_reader__webReader
model: claude-sonnet-4-20250514
---
# Idea-to-Post Expansion Skill
The skill says it will automatically search for and integrate information before and during questioning, but it provides no user-facing warning that user topics may be transmitted to external services. This can expose confidential ideas, internal projects, or personal context through web queries and fetches without informed consent.
The workflow requires collecting real cases, recent experiences, feelings, and emotional reactions as mandatory inputs, but it does not warn users about sensitivity or advise them not to share personal, confidential, or third-party information. In combination with later search/integration steps and generated outputs, this raises privacy leakage and oversharing risks.
The documentation instructs use of a tool name not listed in allowed-tools, creating a mismatch between declared and expected capabilities. Such contradictions can cause fallback behavior, operator confusion, or unsafe substitutions where an agent uses an unintended fetch mechanism to satisfy the instructions.
The documentation states that the skill supports information search integration and advanced search-driven workflows, which expands the effective behavior of the skill beyond the manifest's declared idea-expansion/writing-assistance scope. This can mislead orchestrators or users into granting the skill broader trust or capabilities than intended, increasing the risk of unauthorized external data access or hidden tool use.
This example depicts the skill autonomously searching official docs, reports, and web materials as part of normal execution, even though that capability is not declared in the manifest. Undeclared retrieval behavior is dangerous because it can cause unexpected network access, data exfiltration opportunities, provenance confusion, and user over-trust in generated claims presented as researched facts.
The examples steer users toward Chinese-language output and China-specific platforms without first confirming language or platform preference. While less severe than undeclared tool use, this can cause unwanted disclosure to region-specific ecosystems, mismatched audience targeting, and reduced user control over where content is optimized for publication.
The product-promotion example extends the skill into market and competitor research, which is a materially different capability from simple writing assistance described in the manifest. That scope drift may lead users to expose proprietary product plans or rely on undeclared external research behavior, creating privacy, compliance, and trust-boundary issues.
The document repeatedly steers output toward locale-specific platforms such as WeChat Official Account, Juejin, Zhihu, and Weibo, and the generated content is framed around those ecosystems without stating that this is optional or region-specific. Because the skill guidance implicitly forces a particular locale/platform context rather than offering language or regional choice, it may violate language/locale policy expectations.
The documentation expands the skill from idea expansion into automatic external search/material retrieval, which changes the trust and capability boundary users would reasonably expect from the manifest. Undisclosed capability expansion can cause users to provide sensitive prompts or rely on behavior they were not warned about, especially if external requests transmit user content off-platform.
The examples describe automatic external searching but provide no user-facing warning that network access or prompt-derived data may be transmitted externally. This is dangerous because users may share proprietary ideas, drafts, or business context assuming the skill only performs local writing assistance.
The invocation phrase "Expand idea: [your scattered thought]" is broad enough to match ordinary conversation, increasing the chance of accidental activation. If the skill then begins prolonged questioning or performs search automatically, it could capture more user content than intended or trigger networked behavior without a clear opt-in moment.
This guide provides platform-writing templates and examples entirely in English, including fixed sample titles, hooks, CTAs, and hashtag patterns for platforms that are often used in other languages. Because the document does not offer a language choice or explain that the guidance is intentionally locale-specific, it can be read as enforcing a specific language/locale by default.
This markdown file contains natural-language instructions and question templates solely in English for the idea-to-post skill. Under the policy rule, forcing a specific language without user opt-in is a locale/language policy concern because users are not given a choice of output or interaction language.
The guide hardcodes all example interactions in English and does not instruct the skill to detect or respect the user's preferred language. In a multilingual writing-assistance skill, this can cause the agent to steer users into an unintended language, reducing usability and potentially causing misunderstandings or inaccurate content generation.
The guidance consistently assumes English-language interaction and does not instruct the skill to detect or confirm the user's preferred language. This can lead to miscommunication, accidental mishandling of user intent, or inappropriate prompting in multilingual contexts, which is a safety and reliability issue for an interactive agent skill.
The skill adds web-search and external-validation behavior that is not necessary for the stated purpose of turning a user's ideas into posts. Expanding scope from writing assistance into information gathering can cause unexpected data access, encourage collection of unnecessary external context, and create user-surprising behavior if the skill initiates or relies on search without explicit consent.
The quick-start section claims automatic searching and multi-platform generation beyond the declared skill scope, which can mislead users about what the skill will do and what data handling it performs. While less severe than executable misuse, this creates a specification/documentation mismatch that weakens informed consent and reviewability.
No suspicious patterns detected.