Back to skill

Security audit

看相skill

Security checks for vulnerabilities and agentic risk

Overview

The skill is not a malware payload, but it needs Review because it can route generic human-photo uploads into sensitive appearance-based readings and its README uses an unsafe placeholder install source.

Review before installing. Use only a verified canonical repository or ClawHub-provided package, avoid the placeholder GitHub commands, and prefer project-local installation. Treat results as entertainment only, do not upload other people’s photos without permission, and be aware the skill may make sensitive personality, health, relationship, and fortune claims from appearance.

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)

T08 · Insecure Dependencies

Warning
Location
README.md:26
Finding

Unpinned Placeholder Repository Installed into a Trusted Project Skill Directory

Content
View full analysis

Vulnerability Details

File Location: README.md, lines 26-31
Vulnerability Type: Unverified and mutable supply-chain source
Risk Level: Medium

Vulnerable Code

bash
# 安装到当前项目(在 git 仓库根目录执行)
mkdir -p .claude/skills
git clone https://github.com/yourusername/kanxiang .claude/skills/kanxiang

# 或安装到全局(所有项目都能用)
git clone https://github.com/yourusername/kanxiang ~/.claude/skills/kanxiang

Technical Analysis

The installation instructions clone a repository from the placeholder account yourusername directly into directories trusted by Claude Code. This source conflicts with the repository identity attributed in the same document's footer, which links to timerzz/kanxiang.

The command does not pin an immutable commit, verified tag, checksum, or signed release. Consequently, the content installed by users can differ from the reviewed artifact and can change whenever the remote repository is updated. Installing directly into .claude/skills/ causes Claude Code to treat the downloaded instructions as a trusted Skill.

Although the audited artifact contains no malicious executable payload, this installation process creates a supply-chain boundary through which modified Skill instructions or scripts could be introduced after review.

Attack Path

  1. A user follows the documented installation command.
  2. The placeholder repository resolves to an unintended or attacker-controlled repository, or its content changes after the audit.
  3. The remote content is cloned into .claude/skills/kanxiang or the global ~/.claude/skills/kanxiang directory.
  4. Claude Code discovers and loads the downloaded Skill.
  5. Malicious instructions or scripts supplied by the remote repository execute through the Agent's available tools and permissions.
  6. If globally installed, the altered Skill can affect later sessions across multiple projects.

Impact Assessment

The immediate scope is the user's Claude C ...[truncated 641 chars]

Remediation
View remediation

Remediation Suggestions

  1. Replace https://github.com/yourusername/kanxiang with the verified canonical repository URL.
  2. Ensure the installation URL matches the repository identity used elsewhere in the documentation.
  3. Pin installation to a reviewed immutable commit hash or cryptographically signed release.
  4. Publish and verify a checksum for release archives before installation.
  5. Recommend project-local installation by default; document the expanded trust scope of global installation.
  6. Advise users to inspect SKILL.md and any executable files before enabling the Skill.
  7. Protect release tags and require reviewed, signed changes for published Skill artifacts.

T08 · Insecure Dependencies

Warning
Location
README_EN.md:26
Finding

Unpinned Placeholder Repository Installed into a Trusted Global or Project Skill Directory

Content
View full analysis

Vulnerability Details

File Location: README_EN.md, lines 26-31
Vulnerability Type: Unverified and mutable supply-chain source
Risk Level: Medium

Vulnerable Code

bash
# Install to current project (execute in git repository root)
mkdir -p .claude/skills
git clone https://github.com/yourusername/kanxiang .claude/skills/kanxiang

# Or install globally (available for all projects)
git clone https://github.com/yourusername/kanxiang ~/.claude/skills/kanxiang

Technical Analysis

The English installation instructions direct users to clone an unverified placeholder repository into directories from which Claude Code loads trusted Skills. The URL uses yourusername, while the document footer attributes the project to timerzz/kanxiang, creating an inconsistent repository identity.

No immutable commit, signed tag, checksum, or verified release is specified. The effective Skill installed by a user therefore depends on mutable remote content rather than the artifact covered by this audit. A repository owner or attacker who controls the referenced source could add hostile instructions or executable files after the current review.

The global installation option increases exposure because the downloaded Skill becomes available across all projects using that account.

Attack Path

  1. A user copies one of the installation commands from README_EN.md.
  2. Git retrieves mutable content from the placeholder or subsequently compromised repository.
  3. The content is installed under a trusted project-level or global Skill path.
  4. Claude Code automatically discovers the downloaded Skill.
  5. Attacker-controlled Skill instructions can invoke available Agent tools or direct execution of newly introduced scripts.
  6. A global installation can continue influencing future sessions in unrelated repositories until detected and removed.

Impact Assessment

Successful exploitation could operate with the p ...[truncated 551 chars]

Remediation
View remediation

Remediation Suggestions

  1. Use one verified canonical repository URL consistently across all documentation.
  2. Remove the yourusername placeholder before publication.
  3. Pin the documented installation to an immutable reviewed commit or signed release.
  4. Provide release checksums and instructions for validating downloaded content.
  5. Prefer project-local installation and clearly warn that global installation expands the trust boundary.
  6. Require code review and protected release workflows for changes to Skill instructions and executable files.
  7. Tell users to review downloaded content before allowing Claude Code to load it.
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Rogue AgentSelf-Modification, Session Persistence
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (18)

Vague Triggers

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The trigger keywords include broad phrases like photo-analysis style requests that are not specific to physiognomy, which can cause this skill to activate for ordinary image requests involving faces, hands, or bodies. In context, that increases the chance of unsolicited biometric/personality/health-style inference from sensitive images, making accidental invocation materially risky.

Content

No source excerpt is available for this finding.

Missing User Warnings

High
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill asks users to upload face, hand, body, and bone-structure photos, which are highly sensitive biometric or personal images, but it does not warn about privacy risks before collection and processing. Because the skill also derives character, health, and life-outcome style inferences from appearance, the absence of a privacy notice and consent guard makes the context more dangerous than ordinary image analysis.

Content

No source excerpt is available for this finding.

Session Persistence

Medium
Category
Rogue Agent
Confidence
60% confidence
Finding

Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Content

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

bash
# 安装到当前项目(在 git 仓库根目录执行)
mkdir -p .claude/skills
git clone https://github.com/yourusername/kanxiang .claude/skills/kanxiang

# 或安装到全局(所有项目都能用)

Session Persistence

Medium
Category
Rogue Agent
Confidence
60% confidence
Finding

Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Content

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

bash
# 安装到当前项目(在 git 仓库根目录执行)
mkdir -p .claude/skills
git clone https://github.com/yourusername/kanxiang .claude/skills/kanxiang

# 或安装到全局(所有项目都能用)

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The trigger examples are broad enough that ordinary user messages such as asking to analyze a photo could invoke the skill without clear user intent boundaries. In this skill's context, that increases the chance of unintended processing of sensitive biometric images and generation of physiognomy-style inferences from them.

Content

No source excerpt is available for this finding.

Skill Enumeration

Medium
Category
Agent Snooping
Confidence
85% 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 · README_EN.md (reported line 23)May include surrounding context.

Installation

Claude Code

Important: Claude Code looks for skills in .claude/skills/ from the git repository root. Execute in the correct location.

bash
# Install to current project (execute in git repository root)

Skill Enumeration

Medium
Category
Agent Snooping
Confidence
85% 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 · README.md (reported line 27)May include surrounding context.

bash
# Install to current project (execute in git repository root)
mkdir -p .claude/skills
git clone https://github.com/yourusername/kanxiang .claude/skills/kanxiang

# Or install globally (available for all projects)

Skill Enumeration

Medium
Category
Agent Snooping
Confidence
85% 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 · README_EN.md (reported line 27)May include surrounding context.

bash
# Install to current project (execute in git repository root)
mkdir -p .claude/skills
git clone https://github.com/yourusername/kanxiang .claude/skills/kanxiang

# Or install globally (available for all projects)

Session Persistence

Medium
Category
Rogue Agent
Confidence
60% confidence
Finding

Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Content

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

bash
# Install to current project (execute in git repository root)
mkdir -p .claude/skills
git clone https://github.com/yourusername/kanxiang .claude/skills/kanxiang

# Or install globally (available for all projects)

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The README says users can 'simply upload a photo directly' and that the skill will automatically recognize the photo type and perform analysis. That creates an overly broad activation pattern for a vision-capable agent, increasing the chance the skill runs on ordinary image uploads without explicit user consent for physiognomy analysis, which is especially sensitive because it infers personality, health, and fortune from appearance.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The example trigger 'Help me analyze this photo' is generic and overlaps with common requests for benign image description or troubleshooting. In an agent environment, such broad phrasing can cause the skill to intercept unrelated image tasks and produce sensitive physiognomy-style inferences the user did not specifically request.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The example invocation '帮我分析一下这张照片' is overly ambiguous and teaches the system that a generic photo-analysis request should route into a physiognomy skill. That can misclassify normal image-assistance requests and lead to sensitive inferential analysis of a person's appearance without clear user intent.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

This is a true policy/safety vulnerability because the content makes appearance-based value judgments and gender-specific prescriptions such as linking facial traits to marital worth, fertility, personality, and social roles. In the context of a skill that analyzes uploaded body-part images and produces 'professional analysis,' these stereotypes are operationalized into user-facing outputs, increasing the likelihood of discriminatory, harmful, or demeaning conclusions.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The entire reference file is written only in Chinese and does not indicate that other language options are available or that Chinese is required for a specific justified regional context. Under the stated policy, forcing a specific language without user opt-in is a natural-language policy violation.

Content

No source excerpt is available for this finding.

Vague Triggers

Low
Category
Not specified by scanner
Confidence
83% confidence
Finding

This example uses a generic phrase for analyzing a photo, which may cause the skill to activate in situations where the user only wanted a general image description rather than physiognomy analysis. Because the skill infers personality, career fit, health, and fortune from appearance, accidental invocation can produce intrusive or inappropriate sensitive inferences.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Low
Category
Not specified by scanner
Confidence
78% confidence
Finding

The documentation explicitly states that uploaded photos and analysis results are not saved. In the provided artifact, only README content is available and there is no implementation showing how data handling actually works, so this operational claim is unsupported and may mislead users about retention behavior.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
97% confidence
Finding

This markdown file presents all instructions and reference material exclusively in Chinese, and there is no natural-language indication that users may choose another language or that the skill is intended only for a Chinese-language audience. Under the stated policy, forcing a specific language without opt-in is a locale/language policy concern.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
92% confidence
Finding

SQP-3 applies to all file types and covers language or locale policy violations. This file presents all guidance exclusively in Chinese and does not provide any user opt-in, alternative language option, or justification for a Chinese-only constraint.

Content

No source excerpt is available for this finding.