Back to skill

Security audit

Builder Site institucional

Security checks for vulnerabilities and agentic risk

Overview

The skill is a coherent React site builder, but it includes an unsafe password UI template and under-scoped file, dependency, and third-party asset behaviors that users should review before installing.

Install only if you are comfortable with the agent inspecting a clearly chosen project-materials folder, creating files, and installing npm packages. Do not use the bundled password-confirmation component with real passwords or secrets, and review any external assets, 21st.dev retrieval, stock images, and dependency additions before shipping a generated site.

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

Warning
Location
references/components/assistance_password_comparation.txt:18
Finding

Plaintext Password Disclosure in Client-Side Reference Component

Content
View full analysis
, ) => { if ( confirmPassword.length >= password.length && e.target.value.length > confirmPassword.length ) { setShake(true) } else { setConfirmPassword(e.target.value) } } useEffect(() => { if (shake) { const timer = setTimeout(() => setShake(false), 500) return () => clearTimeout(timer) } }, [shake]) const getLetterStatus = (letter: string, index: number) => { if (!confirmPassword[index]) return '' return confirmPassword[index] === letter ? 'bg-green-500/20' : 'bg-red-500/20' } const passwordsMatch = password === confirmPassword ``` The original password is then rendered directly: ```tsx → {password} ``` ### Technical Analysis The component accepts the authoritative password as a client-side React prop, uses it for character-by-character comparison, and explicitly renders it as visible text. Consequently, a real password supplied to this component would be available in the page DOM, React component state or props, browser developer tools, memory snapshots, screenshots, and screen recordings. Using `type="password"` only masks the confirmation input. It does not protect the original password passed through the `password` prop or the plaintext value rendered by the `` element. The Skill primarily creates public institutional websites. A password-confirmation component is not required for that fun ...[truncated 1203 chars]
Remediation
View remediation
Confirm your password ``` 4. For registration or password-change forms, collect the new password and confirmation value from two masked inputs and compare only those two user-entered values. 5. Send passwords only over TLS to a trusted backend. 6. On the backend, process passwords using a dedicated password-hashing algorithm such as Argon2id, scrypt, or bcrypt with appropriate parameters and unique salts. 7. Never log passwords or expose them through diagnostics, analytics, component props, error messages, or telemetry. 8. Clear password-related client state when the form is submitted, cancelled, or unmounted. 9. Add automated tests that fail if password values are rendered into visible DOM text. ]]>

T08 · Insecure Dependencies

Note
Location
SKILL.md:93
Finding

Unpinned and Mutable Third-Party Dependency Installation

Content
View full analysis
-- --template react-ts cd npm install npm install -D tailwindcss @tailwindcss/vite --legacy-peer-deps npm install framer-motion lucide-react ``` The Skill also permits arbitrary component-specific package installation: ```text Instalar dependências faltantes (`npm install `) ``` ### Technical Analysis The setup workflow resolves `vite@latest` and several dependencies without fixed versions. The actual packages installed can therefore change after the Skill has been reviewed. The generic `npm install ` instruction also permits reference templates to expand the dependency set without a defined allowlist or version policy. NPM packages can execute lifecycle scripts during installation with the privileges of the user running npm. If a mutable dependency, transitive dependency, or newly selected component package becomes compromised, malicious installation logic could execute locally. The use of `--legacy-peer-deps` weakens dependency-resolution safeguards and may produce combinations that were not intended or tested. This does not itself prove package compromise, but it reduces reproducibility and increases supply-chain exposure. ### Attack Path 1. The Agent follows the Skill's project-setup or component-integration instructions. 2. npm resolves `@latest`, unpinned direct packages, or an arbitrary component dependency at execution time. 3. A package owner account, package release, or transitive dependency has been compromised, or an incorrect similarly named package is selected. 4. npm downloads the attacker-controlled package. 5. A package lifecycle script executes during installation with the invoking user's permissions. 6. The maliciou ...[truncated 668 chars]
Remediation
View remediation
`. 5. Review package names carefully to prevent typosquatting and dependency-confusion mistakes. 6. Run `npm audit` or an equivalent software-composition analysis tool before building or delivering the project. 7. Review package provenance, maintainers, release history, integrity metadata, and lifecycle scripts before adding new dependencies. 8. Avoid `--legacy-peer-deps` unless a documented compatibility review establishes that it is necessary. 9. Where feasible, install packages in an isolated, non-privileged environment with restricted access to credentials and sensitive files. 10. Consider disabling lifecycle scripts during initial inspection using `npm install --ignore-scripts`, then explicitly permit only reviewed scripts. ]]>
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
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (54)

Vague Triggers

High
Category
Not specified by scanner
Confidence
95% confidence
Finding

The trigger conditions are extremely broad, including generic requests such as creating a site, landing page, or webpage, and even the presence of visual materials. Overbroad activation increases the chance the skill runs in situations the user did not intend, which can lead to unnecessary file inspection, project creation, and other side effects.

Content

No source excerpt is available for this finding.

Missing User Warnings

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The component visibly renders the supplied password prop in plaintext via → {password}, but the instructions do not warn integrators that sensitive secrets will be exposed in the UI. In a site-building skill context, an agent could mistakenly bind a real user password or secret value to this prop, causing immediate disclosure to end users, screenshots, logs, or recordings.

Content

No source excerpt is available for this finding.

Ssd 3

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The demo passes password='ui.lndev.me' and the component displays that value directly in the interface, normalizing a pattern where the 'password' input is intentionally shown in cleartext. This is dangerous because it encourages insecure reuse in production flows, where real credentials or secret tokens could be exposed in the browser, captured in telemetry, or leaked through shoulder surfing and session recordings.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The skill tells the agent to explore all user-provided materials comprehensively but does not require a user-facing warning or consent flow explaining that files across the folder will be inspected. This undermines informed consent and increases the likelihood of accessing sensitive content beyond what the user expected to share for site generation.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The workflow directs recursive enumeration of the entire user folder and reading all images, PDFs, and presentations before code is written. That is broader than necessary for the stated purpose and can cause the agent to inspect unrelated or sensitive files that happen to be colocated with project assets.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

The instructions explicitly require reading all images and extracting text from PDFs/PPTs without any privacy or data-handling warning. These document types often contain confidential business information, and bulk extraction without scoped consent increases the risk of unnecessary exposure of sensitive data.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The skill directs copying the complete project into the user’s folder as part of delivery, but it does not include a warning or confirmation before performing file writes. Unannounced write behavior can overwrite files, create clutter in sensitive directories, or otherwise modify the user environment in ways they did not explicitly approve.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The skill instructs the agent to fetch additional components from an external API (POST /api/prompts) even though the core task is to build a site from client-provided materials. This expands data flow and behavior beyond the user’s supplied assets, creating unnecessary supply-chain and privacy risk if external content is fetched without explicit user consent and validation.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The eval prompts and skill description use very broad trigger language for when the skill should activate, without clear exclusions or routing constraints. In an agent environment, this can cause over-invocation on loosely related requests, increasing the chance of inappropriate file access, unwanted code generation, or accidental handling of user materials that were not intended for this workflow.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The file begins with Portuguese-only instructional content and continues in that language throughout the document. Because the skill does not offer a user language/locale choice or justify a Portuguese-only constraint, it may violate organizational language policy requirements.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The file gives broad, imperative integration instructions to copy and install a component into a codebase without defining clear trust boundaries, approval gates, or when the component is appropriate to use. In an agent skill context, this can cause over-application of unreviewed third-party code and dependencies, increasing supply-chain and unintended-modification risk.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The component embeds a remote Spline scene URL, which causes client browsers to fetch third-party content automatically. This creates privacy, availability, and supply-chain risks because page visitors and developers may unknowingly contact an external service, and the remote asset can change independently of the local codebase.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The component loads an externally hosted image directly from a CDN without warning or validation. This exposes users to unannounced third-party requests and creates reliability and content-integrity issues if the remote image is replaced, removed, or used for tracking.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The instructions are broadly framed as generic integration guidance without clear constraints on when this component workflow should or should not be used, which can cause the agent to inappropriately activate or apply the skill outside its intended scope. In an agent setting, overly broad task matching increases the chance of misfires, unnecessary file modifications, dependency installation, and unintended code insertion into unrelated projects.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

This plain-text skill content describes a general task to integrate an existing React component, but it does not define clear activation phrases, boundaries, or exclusion conditions. The broad framing could overlap with many ordinary component-integration requests and may cause unintended invocation because there is no explicit trigger scope or negative examples.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The file gives generic integration instructions that can trigger broad project-modifying behavior without clear preconditions, such as verifying the target stack, user consent, or whether the component is actually needed. In an agent skill context, this increases the chance of unintended file creation, dependency changes, and over-application of the skill when a user request only partially matches the described scenario.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The instructions direct the agent to copy files, create folders, and install dependencies without any warning or explicit approval flow for modifying the user's project. In a code-modifying agent, that can lead to unauthorized or surprising changes, dependency drift, and potential supply-chain exposure if packages are installed automatically.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill explicitly directs copy-pasting component code into fixed directories, which can overwrite or conflict with existing files such as /components/ui/glowing-effect.tsx without any safety check, backup, diff review, or rename guidance. In an automated coding workflow, this can destroy local modifications, introduce incompatible code, or silently replace trusted components.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The integration steps are overly broad and operationally permissive: they instruct generic actions like copying code, installing dependencies, adding assets, and choosing placement in the app without defining safety boundaries, review checkpoints, or scope constraints. In an agent skill context, ambiguous instructions can cause the agent to modify unrelated parts of a project or make assumptions about assets and placement, increasing the risk of unintended changes and unsafe automation.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

Forcing the agent to fill image assets with third-party Unsplash stock images removes user choice and can cause unapproved external content to be inserted into client deliverables. In a site-building skill that processes customer branding materials, this is especially risky because it may override provided assets, create licensing/provenance issues, or misrepresent the client's brand without explicit consent.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The file begins with imperative integration instructions that assume the component should be added whenever encountered, without a narrowly defined trigger or approval boundary. In an agent skill, overly broad activation can cause the assistant to apply codebase-modifying behavior in contexts where the user did not explicitly request this component, increasing the chance of unauthorized changes or dependency drift.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The instructions tell the agent to copy files directly into project directories, creating or overwriting code without warning the user that the codebase will be modified. In a code-generation skill, this is risky because it normalizes silent filesystem changes and can introduce unreviewed third-party code into an existing application.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The dependency installation step instructs adding packages like next, motion, lucide-react, and others without warning that this changes the project's dependency graph and may alter build/runtime behavior. This is especially dangerous because adding framework packages such as next to a non-Next codebase can create conflicts, break builds, or expand the attack surface through unnecessary third-party dependencies.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The file gives broad, imperative integration instructions that are not tightly scoped to a specific project context, encouraging automatic copying of code, dependency installation, and asset substitution without clear trigger boundaries or validation steps. In an agent skill ecosystem, this can cause the skill to activate in situations beyond its intended use and push unreviewed code or dependencies into unrelated projects, increasing supply-chain and integrity risk.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
84% confidence
Finding

The file opens with broad integration instructions that are framed as generally applicable and do not define when the component should or should not be installed. In an agentic workflow, vague activation criteria increase the chance of inappropriate code insertion or unnecessary dependency and file changes in projects where the component is not relevant.

Content

No source excerpt is available for this finding.