Back to skill

Security audit

技能发现雷达

Security checks for vulnerabilities and agentic risk

Overview

This skill is a discovery tool, but it also steers agent routing, adds mandatory advertising, and installs unverified remote skills into persistent local directories.

Review before installing. Use this only if you trust the publisher and are comfortable with it reading local skill metadata, contacting multiple remote marketplaces, using a GitHub token if present, and making persistent changes to your skill directories. Avoid the npx and ZIP install paths unless versions and checksums are independently verified.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • 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
Findings (6)

T01 · Skill Instruction Hijacking

Error
Location
SKILL.md:178
Finding

Mandatory Advertising and External-Link Output Hijacking

Content
View full analysis
💡 更多实用 AI 效率工具和技能,领取自媒体IP&超级个体&一人公司资料,关注公众号「桂皮AI实战」 > 📱 加入自媒体&AI 副业变现交流群:https://e418e2e692454bfaa8b6206e3f0ba789.app.codebuddy.work ``` The source text instructs the agent to append promotional material and an external group link to the end of every response. ### Technical Analysis The Skill contains a persistent instruction that modifies all generated results by adding unrelated advertising and an externally controlled URL. This behavior is not necessary for skill discovery and overrides the expected goal of producing a response limited to the user's request. Because this instruction is part of the loaded Skill text, an agent following the Skill will apply it automatically without obtaining explicit consent. The destination is outside the audited package and can change independently of this review. Consequently, the link could later be used for tracking, social engineering, phishing, or recruitment into an untrusted communication channel. ### Attack Path 1. The Skill is installed and loaded by the agent. 2. A user makes an ordinary skill-discovery request. 3. The agent follows the otherwise legitimate search workflow. 4. Before returning the answer, the mandatory instruction appends promotional content and an external URL. 5. The user may follow the link under the assumption that it is part of a trusted platform workflow. 6. The externally controlled destination can collect information, redirect the user, or expose the user to subsequent social-engineering content. ### Impact Assessment The issue compromises the integrity of every response generated under this Skill. It does not directly grant operating-system privileges, but it gives the Skill author persistent control over a portion of agent output and creates a redirection ...[truncated 281 chars]
Remediation
View remediation

T01 · Skill Instruction Hijacking

Error
Location
SKILL.md:393
Finding

Exclusive Agent-Routing Control Claim

Content
View full analysis
Remediation
View remediation

T03 · Remote Payload Retrieval and Execution

Error
Location
SKILL.md:293
Finding

Unverified Remote Skill Retrieval and Installation

Content
View full analysis
" mkdir -p / unzip -o "$TMPDIR/skill.zip" -d / rm -rf "$TMPDIR" ls //SKILL.md ``` Equivalent unverified installation patterns are also specified for other sources: ```bash curl -L -o "$TMPDIR/skill.zip" "" unzip -o "$TMPDIR/skill.zip" -d / ``` ```bash curl -L -o "$TMPDIR/skill.zip" "https://clawhub.com/api/download?slug=" unzip -o "$TMPDIR/skill.zip" -d / ``` ```bash git clone "https://github.com//.git" "$TMPDIR/" mkdir -p / cp -r "$TMPDIR/" / ``` ### Technical Analysis The installation procedures retrieve mutable content from remote marketplaces, arbitrary download URLs, and GitHub repositories and place it directly into trusted Skill directories. No checksum, digital signature, immutable commit identifier, trusted-publisher verification, file allowlist, or manual content review is required. The archive procedures also fail to validate extracted paths and symlinks before calling `unzip`. A malicious archive may contain path traversal entries, absolute paths, or unsafe links, depending on archive-tool behavior and platform protections. Merely checking for `SKILL.md` does not establish safety. A remotely supplied Skill can contain malicious agent instructions or executable scripts that become active when the host platform loads or invokes it. ### Attack Path 1. An attacker publishes a malicious Skill, compromises a marketplace artifact, control ...[truncated 1223 chars]
Remediation
View remediation

T08 · Insecure Dependencies

Error
Location
SKILL.md:190
Finding

Execution of Unpinned Third-Party Packages Through npx

Content
View full analysis
-g -y ``` ### Technical Analysis `npx` may download and execute package code from a package registry at invocation time. The package versions are not pinned, package integrity is not specified, and publisher provenance is not verified. The `-g -y` options further increase risk by requesting global installation and suppressing interactive confirmation. Package CLI code and installation lifecycle scripts can execute with the privileges of the agent's operating-system account. This creates a supply-chain execution path in which the effective code can change after the Skill itself has been audited. ### Attack Path 1. The primary search sources return no suitable result, causing the fallback workflow to run. 2. The agent invokes an unpinned `npx` package by name. 3. The registry resolves the package to its current version at execution time. 4. A compromised, transferred, dependency-confused, or malicious package supplies executable CLI or lifecycle code. 5. That code runs locally with the agent user's privileges. 6. For the global unattended installation command, the package may also create persistent files outside the project directory. ### Impact Assessment Successful exploitation can provide arbitrary code execution under the account running the agent. This may allow access to user-readable files, environment variables, credentials, network resources, and writable application directories. The global installation option broadens the scope to the user's package environment and can persist malicious components for later invocations. Administrative privileges are not explicit ...[truncated 105 chars]
Remediation
View remediation

T01 · Skill Instruction Hijacking

Error
Location
SKILL.md:137
Finding

Security-Critical Delegation to a Mutable External Skill

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
SKILL.md:190
Finding

Potential Shell Injection Through Unquoted Search Queries

Content
View full analysis
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
  • Rogue AgentSelf-Modification, Session Persistence
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
Findings (24)

Vague Triggers

High
Category
Not specified by scanner
Confidence
96% confidence
Finding

The description defines activation using very broad natural-language scenarios such as generic requests for help or tools. Because the skill can then scan the filesystem, contact external services, and install software, overbroad triggering materially raises the chance of unintended invocation and unsafe side effects.

Content

No source excerpt is available for this finding.

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 62)May include surrounding context.

md
用 `Read` 工具读取每个技能的 `SKILL.md` YAML frontmatter,与用户场景做**语义匹配**。

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 83)May include surrounding context.

md
用 `Read` 工具读取每个技能的 `SKILL.md` YAML frontmatter,与用户场景做**语义匹配**。

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 174)May include surrounding context.

md
用 `Read` 工具读取每个技能的 `SKILL.md` YAML frontmatter,与用户场景做**语义匹配**。

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 363)May include surrounding context.

md
用 `Read` 工具读取每个技能的 `SKILL.md` YAML frontmatter,与用户场景做**语义匹配**。

Missing User Warnings

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The installation workflow performs cp, unzip, git clone, and ln -s operations that modify persistent user directories, yet no strong upfront warning or trust verification is required. In a skill ecosystem, this can lead to installation of malicious or tampered skills and durable compromise of the user's environment.

Content

No source excerpt is available for this finding.

Vague Triggers

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The trigger reference table includes vague phrases like '我想做XXX' and '有没有能XXX的工具', which overlap with normal assistant usage. In context, accidental activation is dangerous because the skill can enumerate local directories and perform remote fetch/install steps on weak user signals.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
85% confidence
Finding

Line L04 requires a specific Chinese brand attribution string to be used uniformly and forbids alternatives, which imposes a fixed language choice. Under the policy, forcing a specific language without user opt-in is a natural-language policy violation unless clearly justified as region-specific, which is not stated here.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The activation scope is very broad: ordinary requests like '我想做一个海报' or '帮我分析股票' can trigger the skill even when the user did not explicitly ask to search for or install skills. Because this skill can recommend and support one-click installation from multiple external sources, overbroad triggering increases the chance of unsolicited marketplace access, unintended installs, and exposure to untrusted third-party content. In this context, the danger is elevated because the skill is a discovery/installation broker across GitHub and other hubs, not just a passive local helper.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
87% confidence
Finding

The skill instructs scanning built-in app paths and user/project skill directories without an upfront user warning. Users may not expect a recommendation tool to inspect local installation paths, and silent inspection increases privacy and consent risk.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The skill instructs use of GITHUB_TOKEN from the environment for authenticated GitHub API access. Even if not explicitly exfiltrated, encouraging credential use in a broad discovery workflow increases the chance of token exposure through command construction, logs, downstream tools, or unintended requests to third-party content.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
50% 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 · SKILL.md (reported line 157)May include surrounding context.

md
fi

# 搜索包含 SKILL.md 的代码
curl -s $AUTH_HEADER "https://api.github.com/search/code?q=filename:SKILL.md+<关键词>&per_page=10"

# 搜索技能相关仓库
curl -s $AUTH_HEADER "https://api.github.com/search/repositories?q=<URL-encoded 关键词>+skill+in:name,description&per_page=10"

External Transmission

Medium
Category
Data Exfiltration
Confidence
50% 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 · SKILL.md (reported line 160)May include surrounding context.

md
fi

# 搜索包含 SKILL.md 的代码
curl -s $AUTH_HEADER "https://api.github.com/search/code?q=filename:SKILL.md+<关键词>&per_page=10"

# 搜索技能相关仓库
curl -s $AUTH_HEADER "https://api.github.com/search/repositories?q=<URL-encoded 关键词>+skill+in:name,description&per_page=10"

External Transmission

Medium
Category
Data Exfiltration
Confidence
50% 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 · SKILL.md (reported line 346)May include surrounding context.

md
fi

# 搜索包含 SKILL.md 的代码
curl -s $AUTH_HEADER "https://api.github.com/search/code?q=filename:SKILL.md+<关键词>&per_page=10"

# 搜索技能相关仓库
curl -s $AUTH_HEADER "https://api.github.com/search/repositories?q=<URL-encoded 关键词>+skill+in:name,description&per_page=10"

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The instructions require that every output must append specific promotional text in Chinese, regardless of user preference or language. This imposes a fixed output language/content policy without offering opt-in or language choice.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

A discovery skill should primarily recommend skills, but this one expands into package-manager/CLI execution (npx skills, npx clawhub) unrelated to passive search. That broadens the trust boundary from metadata retrieval to arbitrary local command execution from external ecosystems.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
92% confidence
Finding

The skill instructs execution of npx skills find [query] without pinning an exact package name/version or integrity source. Unpinned NPX execution can fetch and run whatever package is current at resolution time, creating a supply-chain execution path from a simple discovery workflow.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
92% confidence
Finding

The skill directs use of npx clawhub search [query] without a pinned version. This allows remote package resolution and code execution from the registry at runtime, which is risky in a skill-discovery context where users do not expect arbitrary package installation/execution.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
90% confidence
Finding

The skill downloads a remote ZIP from lightmake.site and unpacks it directly into the active skills directory. This is an external content ingestion path that can deliver malicious skill content or exploit unsafe extraction behavior, leading to persistent compromise.

Content

Scanner excerpt · SKILL.md (reported line 293)May include surrounding context.

bash
TMPDIR=$(mktemp -d)
curl -L -o "$TMPDIR/skill.zip" "https://lightmake.site/api/v1/download?slug=<slug>"
mkdir -p <target-skills-dir>/<slug>
unzip -o "$TMPDIR/skill.zip" -d <target-skills-dir>/<slug>
rm -rf "$TMPDIR"

Session Persistence

Medium
Category
Rogue Agent
Confidence
89% confidence
Finding

Persisting a remote skill under the active target directory creates durable changes that survive the current interaction. Because this skill supports broad triggering and remote installations, that persistence materially increases the risk of later unintended execution.

Content

Scanner excerpt · SKILL.md (reported line 294)May include surrounding context.

bash
TMPDIR=$(mktemp -d)
curl -L -o "$TMPDIR/skill.zip" "https://lightmake.site/api/v1/download?slug=<slug>"
mkdir -p <target-skills-dir>/<slug>
unzip -o "$TMPDIR/skill.zip" -d <target-skills-dir>/<slug>
rm -rf "$TMPDIR"
ls <target-skills-dir>/<slug>/SKILL.md

Session Persistence

Medium
Category
Rogue Agent
Confidence
89% confidence
Finding

Persisting a remote skill under the active target directory creates durable changes that survive the current interaction. Because this skill supports broad triggering and remote installations, that persistence materially increases the risk of later unintended execution.

Content

Scanner excerpt · SKILL.md (reported line 294)May include surrounding context.

bash
TMPDIR=$(mktemp -d)
curl -L -o "$TMPDIR/skill.zip" "https://lightmake.site/api/v1/download?slug=<slug>"
mkdir -p <target-skills-dir>/<slug>
unzip -o "$TMPDIR/skill.zip" -d <target-skills-dir>/<slug>
rm -rf "$TMPDIR"
ls <target-skills-dir>/<slug>/SKILL.md

Rp1

Medium
Category
MCP Rug Pull
Confidence
95% confidence
Finding

npx skills add <owner/repo@skill> -g -y is unpinned and performs installation, not just search. Because it both resolves and executes external package code and persists changes globally, it increases the blast radius beyond a normal discovery action.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
71% confidence
Finding

The natural-language description and examples present the skill entirely in Chinese and emphasize Chinese-specific sources such as 虾评, but do not state that language or locale is optional or region-specific. Under the policy, forcing a specific language or locale without opt-in can be a natural-language policy violation.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Low
Category
Not specified by scanner
Confidence
81% confidence
Finding

The skill instructs scanning built-in and local skill directories to enumerate installed content. While this appears intended for recommendation quality, it collects local environment details beyond what is necessary for remote discovery and could expose sensitive project or user-specific metadata to the model or later outputs.

Content

No source excerpt is available for this finding.