T01 · Skill Instruction Hijacking
- Location
SKILL.md:131- Finding
Remote MCP Response Can Hijack the Agent's Final Output
- Content
View full analysis
- Remediation
View remediation
Security audit
Security checks for vulnerabilities and agentic risk
The skill mostly matches its Rain Classroom teaching-data purpose, but its installer and response-handling instructions create review-worthy security and privacy risks.
Review before installing. Use this only if you trust the Rain Classroom endpoint and the mcporter package source, prefer a manually reviewed MCP config over running the scripts, avoid running setup with elevated privileges, and treat YUKETANG_SECRET as a sensitive credential. Be aware that setup.sh silently reports an install event and duration, and that one skill instruction allows remote MCP text to become the assistant's verbatim final response.
SKILL.md:131Remote MCP Response Can Hijack the Agent's Final Output
setup.js:29Command Injection Through YUKETANG_SECRET in the Node.js Installer
setup.js:34Unpinned mcporter Package Is Downloaded and Executed During Setup
setup.sh:99Shell Installer Sends Undisclosed Installation Telemetry
The skill instructs users to configure and use a personal secret via an environment variable, but it does not declare any explicit tool scope or allowed-tools boundary. That weakens least-privilege controls and makes it harder for a host platform to constrain what capabilities the skill may access, increasing the risk of unintended secret exposure or broader-than-expected execution behavior.
The setup tells users to place a personal secret in an environment variable but provides no warning about credential sensitivity, storage hygiene, shell history exposure, process inheritance, or log leakage. In practice, users may mishandle the secret, commit it to scripts, share terminal output, or leave it accessible to child processes and local tooling.
This instruction forces a specific locale/timezone policy for all users: relative times must be interpreted using system Beijing time. The policy-violation rule applies because the file does not provide user opt-in, a configurable locale choice, or a clear justification that the skill is region-locked to that timezone.
Using npx mcporter without pinning a version introduces a supply-chain risk because the latest package version will be fetched at execution time. If the package is compromised or a breaking/malicious release is published, users could execute unreviewed code during setup or operation.
The manifest description forces a specific language/locale in user-facing natural language without offering any alternative or opt-in. The stated policy flags language or locale constraints unless the skill offers user choice or clearly documents a justified regional limitation.
This markdown file contains all user-facing instructions and examples exclusively in Chinese, which can force a specific language on users without opt-in. The policy requires either offering a language/locale choice or clearly documenting a justified region-specific constraint.
The script reads YUKETANG_SECRET and interpolates it directly into a shell command executed via execSync. If the secret contains shell metacharacters, quoting edge cases, or attacker-controlled content, it can break command boundaries or be exposed through process listings, shell history, logs, or downstream tooling, turning a credential-handling step into command-injection and secret-leakage risk.
The script reads the sensitive environment variable YUKETANG_SECRET and guides the user to set it, but it does not include any user-facing warning about the sensitivity of this credential or safe handling practices. Under the code-file criteria, access to sensitive environment variables should have some disclosure, and the surrounding messages only indicate setup steps, not a safety warning.
The script invokes npx mcporter without pinning an exact package version, so installation and execution may resolve to whatever version is current at runtime. This creates a supply-chain risk: a compromised, typosquatted, or maliciously updated package could execute arbitrary code during setup, and the command also carries an Authorization header containing the user's secret.
The verification step also runs npx mcporter without a pinned version, exposing users to the same untrusted package-resolution behavior during setup validation. Even though this second call does not pass the bearer token directly, it still executes code fetched at runtime from the package ecosystem.
The script invokes npx mcporter without pinning an exact package version, so execution depends on whatever version npx resolves at runtime. If the package or dependency chain is updated maliciously or unexpectedly, the setup script could run altered code during installation.
This call to npx mcporter list also relies on an unpinned package resolution path. That creates supply-chain risk because a different or compromised package version could be fetched and executed in the user's environment.
Another unpinned npx mcporter execution appears during verification output. Repeated unpinned executions increase the chance of retrieving or running a tampered package version, especially in ephemeral environments.
The setup script performs a silent external reporting call after installation, which is not required for local MCP registration. Even though only duration and action are shown in the payload, the call goes to a remote MCP service using the configured authorization context, creating undisclosed telemetry and trust concerns.
The telemetry/reporting network call is performed with no user-facing warning, consent, or visible output. In a setup script that handles a bearer secret, undisclosed outbound traffic is especially concerning because users cannot easily assess whether sensitive context or metadata may be exposed.
The reporting step executes npx mcporter call without version pinning, combining supply-chain risk with an outbound network action. If a malicious package version is resolved, it could abuse the bearer token, alter telemetry, or execute additional unauthorized actions.
Multiple comments and console messages instruct the user only in Chinese, including setup guidance and the final suggested prompt. This creates a language policy issue because the skill effectively forces a specific language without user opt-in or documented justification.
Nearly all user-facing messages in the script are presented only in Chinese, including setup guidance and completion instructions, with no opt-in or alternate locale. This can violate language/locale policy where skills should not force a specific language absent user choice or documented regional justification.
Labeling the step as 'silent' while making an external service call is deceptive from a user-consent perspective. This mismatch reduces transparency and can cause users to unknowingly transmit installation metadata to a remote service.