Back to skill

Security audit

Youtube Thumbnail Design

Security checks for vulnerabilities and agentic risk

Overview

This skill provides relevant thumbnail guidance, but it asks users to run mutable remote installers and unpinned dependency commands that can execute code on their machine.

Review this skill before installing. The thumbnail advice itself is ordinary, but do not run the curl-to-shell or npx commands from a normal workstation with sensitive credentials. Prefer a pinned, manually verified installer, inspect the downloaded script or binary first, run it without sudo in a restricted environment, and log in only after you trust the installed CLI.

Vulnerability Patterns
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • 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
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
Findings (2)

T03 · Remote Payload Retrieval and Execution

Error
Location
SKILL.md:14
Finding
Remote installer is downloaded and executed without prior inspection<![CDATA[ ## Vulnerability Details **File Location**: `SKILL.md`, line 14 **Vulnerability Type**: Remote payload retrieval and execution through a `curl | sh` pipeline **Risk Level**: High ### Vulnerable Code ```bash curl -fsSL https://cli.inference.sh | sh && infsh login ``` ### Technical Analysis This instruction downloads a shell script from an external URL and immediately passes it to `sh`. The effective code being executed is not contained in the audited project and can change after the Skill has been reviewed. Although the accompanying installation note states that the script selects a platform-specific binary and verifies its SHA-256 checksum, the checksum validation is itself performed by the untrusted remote script. The pipeline does not independently pin or verify the installer script before execution. HTTPS protects the connection in transit but does not protect against compromise of the hosting service, DNS or certificate infrastructure, deployment pipeline, or publisher account. The thumbnail-generation task does not inherently require arbitrary remote shell execution. Installing a command-line client may be a legitimate prerequisite, but executing a mutable installation script directly from the network exceeds the minimum privilege and trust needed for that setup. ### Attack Path 1. An attacker compromises `cli.inference.sh`, its deployment pipeline, hosting account, or another trusted distribution component. 2. The attacker replaces or modifies the returned installer script. 3. A user or agent follows the Quick Start instruction. 4. `curl` retrieves the attacker-controlled response and streams it directly into `sh`. 5. The payload executes immediately with all permissions of the current user, before `infsh login` is run. 6. The payload can access user-readable data, alter files, steal environment credentials, download additional components, or establish persistence where current-user permissions allow it. ### Impact Assessment Successful ...[truncated 747 chars]
Remediation
<![CDATA[ ## Remediation Suggestions 1. Remove the `curl | sh` pipeline from the recommended installation path. 2. Direct users to download a version-pinned release artifact separately. 3. Publish checksums through an independently trusted, versioned release channel and require verification before execution. 4. Prefer signed release artifacts and verify the signature against a documented publisher key. 5. Show a transparent manual installation sequence, for example: - Download a specific version to a local file. - Verify its cryptographic digest or signature. - Inspect or extract the artifact. - Install it into a user-owned directory without administrative privileges. 6. If an installer script remains available, pin its immutable version and digest rather than retrieving a mutable URL. 7. Keep installation separate from authentication so users can validate the installed binary before entering credentials. 8. Explicitly warn users not to execute the installer as root or with `sudo`. ]]>

T08 · Insecure Dependencies

Warning
Location
SKILL.md:244
Finding
Unpinned executable and Skill dependencies are installed through npx<![CDATA[ ## Vulnerability Details **File Location**: `SKILL.md`, lines 244-246 **Vulnerability Type**: Unpinned third-party executable and mutable Skill dependencies **Risk Level**: Medium ### Vulnerable Code ```bash npx skills add inference-sh/skills@ai-image-generation npx skills add inference-sh/skills@image-upscaling npx skills add inference-sh/skills@prompt-engineering ``` ### Technical Analysis The commands invoke `npx skills` without pinning the `skills` package to an exact version or integrity value. Depending on the local environment and package-manager behavior, `npx` can retrieve and execute a package-provided CLI from the package registry. A future or compromised version would therefore run during installation with the current user's permissions. The added Skill references also use named, mutable identifiers rather than immutable commit hashes or content digests. Their contents can consequently change after this audit. Since Skills may affect agent instructions and tool usage, installing a modified Skill can introduce harmful behavior even when it does not include a conventional executable. These dependencies are listed as optional related Skills and are not necessary to perform the declared thumbnail-design workflow. Recommending their immediate installation therefore expands the supply-chain trust boundary beyond the minimum required functionality. ### Attack Path 1. An attacker compromises the registry package used by `npx skills`, its maintainer account, or the source behind one of the referenced Skill identifiers. 2. The attacker publishes a malicious package version or modifies content reachable through a mutable Skill reference. 3. A user runs one of the documented commands. 4. `npx` resolves and may execute the unpinned CLI package. 5. The CLI installs content that has not been pinned to the version reviewed by the user. 6. Malicious package code can execute with current-user privileges, or malicious Skill instructions can affect lat ...[truncated 713 chars]
Remediation
<![CDATA[ ## Remediation Suggestions 1. Pin the `skills` CLI to an exact reviewed version rather than invoking an unresolved package name. 2. Use package-manager lockfiles and integrity hashes where installation is part of an automated environment. 3. Pin each referenced Skill to an immutable commit, release digest, or other content-addressed identifier. 4. Review the complete contents, allowed tools, and transitive installation instructions of each Skill before enabling it. 5. Do not automatically install optional related Skills as part of the core thumbnail workflow. 6. Run dependency installation in a restricted environment without sensitive credentials and with minimal filesystem and network access. 7. Enable dependency provenance or signature verification where the package registry and Skill distribution mechanism support it. 8. Document the publisher, expected version, integrity value, and update-review procedure for every recommended dependency. ]]>
Vulnerability Patterns
  • 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
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (7)

External Script Fetching

High
Category
Supply Chain
Content
## Quick Start

```bash
curl -fsSL https://cli.inference.sh | sh && infsh login

# Generate a thumbnail
infsh app run falai/flux-dev-lora --input '{
Confidence
99% confidence
Finding
`curl ... | sh` executes a remote script directly in the user's shell without prior inspection, creating a classic arbitrary code execution and supply-chain risk. Even with claims about checksum verification, the bootstrap script itself is still trusted implicitly at execution time, so compromise of the host, transport, or distribution path could lead to full user-level code execution.

Description-Behavior Mismatch

Medium
Confidence
94% confidence
Finding
The manifest presents the skill as thumbnail-design advice, but the content materially drives users into installing and using an external AI generation CLI. This mismatch can mislead reviewers and users about the skill's operational behavior, making risky execution steps seem like harmless design guidance.

Vague Triggers

Medium
Confidence
95% confidence
Finding
The manifest description lists triggers such as "thumbnail design," "thumbnail tips," "video cover image," and especially "ctr optimization," which are broad phrases that may occur in ordinary creative or marketing conversations outside this specific skill context. The file does not provide tighter activation constraints, exclusions, or negative examples to clarify when the skill should or should not be invoked.

Context-Inappropriate Capability

Medium
Confidence
91% confidence
Finding
The skill grants broad Bash capability and includes third-party CLI installation and execution steps that go beyond passive thumbnail-design guidance. This expands the attack surface by encouraging shell execution of externally sourced tooling, increasing the chance of unintended code execution or environment compromise if users follow the instructions.

Rp1

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding
npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Rp1

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding
npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Rp1

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding
npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Static analysis

No suspicious patterns detected.