Back to skill

Security audit

Ielts Review Upload

Security checks for vulnerabilities and agentic risk

Overview

This skill is purpose-built for uploading IELTS review files, but it has review-worthy privacy and credential-handling risks before installation.

Review this skill before installing. It will upload your review data to tuyaya.online, send a stable identifier derived from your machine and local username unless overridden, and may expose a shared API key in links and logs. Avoid using the remote script fallback, set your own explicit IELTS_USER_ID and IELTS_USER_NAME if you proceed, and do not pass optional HTML files unless you are comfortable uploading their full contents.

Vulnerability Patterns
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • 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
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
Findings (4)

T03 · Remote Payload Retrieval and Execution

Error
Location
SKILL.md:57
Finding

Unpinned Remote Shell Script Retrieval and Execution

Content
View full analysis
``` If the script is missing, the Skill instructs the Agent to retrieve it dynamically: ```bash curl -o scripts/sync-review.sh https://raw.githubusercontent.com/dengjiawei1226/ielts-reading-review/main/scripts/sync-review.sh chmod +x scripts/sync-review.sh ``` ### Technical Analysis The Skill retrieves executable shell code from the mutable `main` branch of a personal GitHub repository. It does not pin the download to an immutable commit, validate a cryptographic checksum, verify a digital signature, or inspect the downloaded content before execution. Although the current package includes a local copy of the script, the documented fallback creates a remote code execution channel whenever that file is missing. The effective behavior of the Skill can therefore change after review without any modification to the audited package. This behavior exceeds the minimum privileges required to upload a JSON document. Upload functionality can be implemented entirely by the bundled, reviewed script and does not require downloading replacement executable code at runtime. ### Attack Path 1. An attacker compromises or otherwise gains control of the referenced GitHub account or repository. 2. The attacker modifies `scripts/sync-review.sh` on the `main` branch to include malicious shell commands. 3. The local bundled script is absent, removed, or unavailable. 4. The Agent follows the fallback instructions and downloads the attacker-controlled script. 5. The Agent invokes the downloaded script with Bash during the normal upload workflow. 6. The malicious commands execute with all permissions available to the Agent process. ### Impact Assessment Successful exploitation provides arbitrary command execution under the account running th ...[truncated 504 chars]
Remediation
View remediation

other

Warning
Location
scripts/sync-review.sh:27
Finding

Misleading Anonymity Claim and Unnecessary Host Identity Disclosure

Content
View full analysis
/dev/null; then USER_ID=$(echo -n "$RAW_ID" | shasum -a 256 | cut -c1-16) elif command -v sha256sum &>/dev/null; then USER_ID=$(echo -n "$RAW_ID" | sha256sum | cut -c1-16) else USER_ID=$(echo -n "$RAW_ID" | md5sum 2>/dev/null | cut -c1-16 || echo "$RAW_ID") fi fi USER_NAME="${IELTS_USER_NAME:-$(whoami)}" ``` The derived identifier and raw username are transmitted to the remote service: ```bash RESPONSE=$(curl -s -w "\n%{http_code}" \ -X POST "$API_BASE/skill-review" \ -H "Content-Type: application/json" \ -H "x-api-key: $API_KEY" \ -H "x-user-id: $USER_ID" \ -H "x-user-name: $USER_NAME" \ -d "$PAYLOAD") ``` ### Technical Analysis The documentation characterizes the identifier as anonymous, but the value is a stable fingerprint derived from the machine hostname and local username. It is truncated to 16 hexadecimal characters, reducing collision resistance, while still allowing persistent correlation of uploads. Hashing does not make low-entropy or guessable identifiers anonymous. An operator that can guess likely usernames and hostnames may calculate candidate hashes and compare them with the submitted identifier. More importantly, the script separately transmits the raw output of `whoami` in the `x-user-name` header, directly disclosing the local account name. A stable machine fingerprint and raw operating-system username are not necessary to upload review records. A random application-specific identifier stored locally would provide continuity without deriving identity from host metadata. ### Attack Path 1. A user invokes the Skill to u ...[truncated 1096 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
scripts/sync-review.sh:22
Finding

Hardcoded Shared API Key Exposed in Source and Dashboard URLs

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/sync-review.sh:54
Finding

Unrestricted Optional File Embedding and Upload

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
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (13)

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The documented purpose is limited to uploading review JSON and viewing progress, but the behavior includes collecting system-derived identifiers, embedding a default API key, and returning dashboard URLs containing sensitive access material. This mismatch hides privacy and credential risks from the user and could lead to unintended data disclosure or account access.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
92% confidence
Finding

The skill instructs the agent to read local files and execute shell commands, but it does not declare any tool scope or permissions boundaries. That increases the chance an agent will perform sensitive filesystem or command execution actions without explicit review or least-privilege constraints.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The skill directs the agent to upload local review data and derive machine-linked identifiers without a clear privacy warning or informed consent flow. Users may not realize that personal study records and device-associated metadata are being transmitted to a remote server.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

Including a built-in API key in the skill exposes a credential to anyone who can read the skill and encourages unsafe secret handling. If reused by unauthorized parties, it could enable abuse of the backend service, unauthorized uploads, or wider compromise depending on server-side permissions.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

Deriving a persistent identifier from hostname and username creates cross-session tracking tied to local machine attributes that are unrelated to the core task. Even if hashed, the source inputs are stable and low-entropy enough to create privacy risks and enable correlation of activity across uploads.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The script embeds a default API key and silently transmits review content plus user-derived identifiers to a remote API. Hardcoding credentials in distributed client-side code enables unauthorized reuse of the key, while the lack of prominent disclosure undermines informed consent about what data leaves the machine.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The script derives a stable identifier from local hostname and username and also transmits the system username to a remote service. Even if hashed, this is still host-linked tracking data and exceeds what is necessary for a simple review upload, creating privacy risk and enabling cross-session correlation of a user's machine identity.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The skill claims to upload review JSON, but it also reads arbitrary local HTML files and embeds their full contents into the outbound payload. This broadens the data collection surface beyond the stated purpose and can unintentionally exfiltrate sensitive local content if users pass the wrong file or the workflow encourages attaching generated artifacts containing private data.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
88% confidence
Finding

External transmission is expected for a sync/upload skill, so the network request itself is not inherently malicious. However, in this implementation it becomes security-relevant because the request sends review data, an embedded API key, and host/user-derived identifiers to a third-party endpoint, increasing privacy and credential exposure risk.

Content

Scanner excerpt · scripts/sync-review.sh (reported line 107)May include surrounding context.

sh
# ─── Upload ───
echo "  Uploading to $API_BASE/skill-review ..."
RESPONSE=$(curl -s -w "\n%{http_code}" \
  -X POST "$API_BASE/skill-review" \
  -H "Content-Type: application/json" \
  -H "x-api-key: $API_KEY" \

External Script Fetching

Low
Category
Supply Chain
Confidence
94% confidence
Finding

The skill instructs the agent to fetch and execute a shell script directly from GitHub at runtime, which creates a supply-chain risk. If the remote repository is compromised, the URL changes content, or TLS trust is subverted, the agent may run attacker-controlled code on the local system.

Content

Scanner excerpt · SKILL.md (reported line 70)May include surrounding context.

如果脚本不存在: 从当前 Skill 目录复制,或者从 GitHub 下载:

bash
curl -o scripts/sync-review.sh https://raw.githubusercontent.com/dengjiawei1226/ielts-reading-review/main/scripts/sync-review.sh
chmod +x scripts/sync-review.sh

Context-Inappropriate Capability

Low
Category
Not specified by scanner
Confidence
76% confidence
Finding

The manifest scopes the skill to uploading review JSON to the server and viewing a personal dashboard. Opening a browser or generating a QR code introduces additional device-interaction capabilities that are not declared as part of the skill's purpose.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Low
Category
Not specified by scanner
Confidence
91% confidence
Finding

The code comments downplay sensitivity, but the script sends the actual local system username in an HTTP header to the server. Local account names can reveal a person's name, employer naming conventions, or shared environment details, so mischaracterizing this increases the chance of unnoticed personal data leakage.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
97% confidence
Finding

The script prints its final dashboard guidance entirely in Chinese, which imposes a specific language on all users regardless of locale. The file does not offer an opt-in, fallback language, or justification for the language constraint, so it violates the language/locale policy criterion.

Content

No source excerpt is available for this finding.