Back to skill

Security audit

QR Bridge

Security checks for vulnerabilities and agentic risk

Overview

The skill is purpose-aligned, but it automatically fetches attacker-controlled QR or short-link destinations without clear confirmation or network-safety limits.

Review this skill before installing. It is designed to contact URLs found in QR codes and short links, so do not use it on private login links, payment links, invite links, internal company URLs, or untrusted QR images unless you are comfortable making requests from your machine. Prefer a decode-first workflow where the raw URL is shown before any trace or inspect step runs.

Vulnerability Patterns
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • 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
Findings (2)

T09 · Insecure Skill Coding Practices

Error
Location
SKILL.md:138
Finding

Automatic Fetching of Untrusted QR URLs Enables SSRF and Local Resource Access

Content
View full analysis
&1 # Step 2: Check with browser-like UA curl -sL --max-redirs 10 --connect-timeout 10 \ -H "User-Agent: Mozilla/5.0 (iPhone; CPU iPhone OS 17_0 like Mac OS X) AppleWebKit/605.1.15" \ "URL_HERE" | head -c 5000 # Step 3: Check with WeChat UA (reveals WeChat-gated content) curl -sL --max-redirs 10 --connect-timeout 10 \ -H "User-Agent: Mozilla/5.0 (iPhone; CPU iPhone OS 17_0 like Mac OS X) AppleWebKit/605.1.15 MicroMessenger/8.0.0" \ "URL_HERE" | head -c 5000 ``` The requests are automatically initiated by the following workflow: ```text When the user provides a QR code image, run the full pipeline automatically: 1. DECODE the image 2. If the decoded content is a URL: a. TRACE the redirect chain b. INSPECT the final destination for gates c. DIAGNOSE with full report ``` ### Technical Analysis The decoded QR value is attacker-controlled input. The Skill instructs the Agent to pass that value directly to `curl` without validating its scheme, host, resolved IP address, port, or redirect destinations. Shell quoting prevents ordinary shell metacharacter injection, but it does not make the requested resource safe. An attacker can encode URLs that point to: - Loopback services such as `http://127.0.0.1`; - Private network services such as RFC 1918 addresses; - Link-local or cloud metadata endpoints; - Internal hostnames resolvable only from the victim environment; - Local resources through schemes supported by the installed `curl`; - Public endpoints that redirect to restricted network destinations. The use of `--max-redirs` permits redirect traversal, but every redirect target is ...[truncated 1710 chars]
Remediation
View remediation

T08 · Insecure Dependencies

Warning
Location
scripts/setup.sh:34
Finding

Setup Automatically Installs an Unpinned Python Dependency

Content
View full analysis
/dev/null; then echo " OK: qrcode library available" else echo " Installing qrcode[pil]..." pip3 install "qrcode[pil]" 2>&1 | sed 's/^/ /' fi ``` The Skill documentation also recommends the same installation behavior: ```bash pip3 install "qrcode[pil]" ``` ### Technical Analysis The setup script automatically retrieves and installs the latest version satisfying `qrcode[pil]`, including its transitive dependencies, from whatever package index the local `pip3` configuration selects. No exact version, lock file, package hash, trusted index, or integrity check is specified. Consequently, the effective code installed during setup can change after the Skill itself has been reviewed. Installation also occurs in the active Python environment rather than an isolated project environment. Potential compromise sources include: - A future compromised package or transitive dependency release; - A malicious or misconfigured private package index; - Dependency substitution in an environment with altered pip configuration; - Unexpected breaking or unsafe behavior introduced by later releases. Python package installation may execute build backends or installation-related code with the permissions of the user running setup. ### Attack Path 1. The victim runs `scripts/setup.sh` in an environment where the `qrcode` module is absent. 2. The script invokes the ambient `pip3` executable without user confirmation. 3. Pip resolves the current package and transitive dependency versions from its configured indexes. 4. A compromised package release, dependency, or package source supplies malicious installation or runtime code ...[truncated 663 chars]
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (9)

Tp4

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

The declared description promises end-user link analysis behavior: QR decoding plus redirect tracing and gated-destination inspection. This code chunk instead performs environment setup: compiling a local Swift decoder and ensuring supporting Python libraries are present. While QR decoding setup is related support work, the chunk does not demonstrate the broader declared behaviors around URL tracing or destination analysis. It also exercises an undeclared capability—installing software dependencies—which is operationally significant compared with the description. Therefore the supplied code chunk does not accurately represent the declared purpose.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The declared description suggests a link-analysis tool that goes beyond QR decoding into redirect tracing and gated-destination inspection. The actual code chunk does not perform any network access, URL fetching, redirect following, or destination analysis. Instead, it is a local test script for a QR decoder binary, focused on compile/run validation and error handling. While QR decoding is partially aligned with the description, the broader claimed capabilities are not represented in this code, and the primary purpose of this chunk is automated testing rather than inspecting what a link actually needs.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The README says to use the skill naturally with phrases like "Scan the QR code in this image" and "Where does this short link go?" without defining activation boundaries, exclusions, or a constrained command surface. These are generic requests that could match many ordinary conversations, increasing the risk of unintended invocation.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

This markdown file describes triggering the skill with general phrases like “扫一下这张图里的二维码” and “这个短链最终跳到哪里?”, but it does not define clear trigger boundaries, exclusions, or negative examples. Because the README presents these as natural-language activation patterns for skill use, the lack of specificity could cause unintended invocation when similar everyday requests appear in broader conversations.

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 · SKILL.md (reported line 39)May include surrounding context.

Step 1: Check if binary exists

bash
ls ~/.claude/skills/qr-bridge/scripts/qr-decode 2>/dev/null && echo "COMPILED" || echo "NEED_SETUP"

If NEED_SETUP, run:

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The inspection workflow automatically sends HTTP requests to decoded URLs, including alternate User-Agent probes, without a user-facing warning. QR contents can be sensitive or attacker-controlled, and automatic requests may disclose the user's IP, trigger tracking, hit malicious infrastructure, or interact with single-use/private links.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The 'full pipeline automatically' instruction causes decoded QR content to be fetched and analyzed without pausing for user review. Because QR codes often encode private auth links, payment links, invitation tokens, or internal URLs, this can expose sensitive destinations and cause unintended interactions with external services.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The blog post instructs users to invoke the skill with a very generic phrase like '扫一下这个二维码', which is ordinary conversational language likely to appear in many unrelated interactions involving images or screenshots. In an auto-activation skill system, broad triggers can cause the skill to run unexpectedly on untrusted content, increasing the chance of unintended network access, QR decoding, redirect tracing, or misclassification of user intent.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
93% confidence
Finding

The QR generation command writes a file to disk, defaulting to ~/Desktop/qr-output.png, without requiring explicit user acknowledgement. Unexpected file creation can overwrite user expectations, leak sensitive encoded content to a predictable location, and create privacy issues on shared systems.

Content

No source excerpt is available for this finding.