Back to skill

Security audit

Video Ad Specs

Security checks for vulnerabilities and agentic risk

Overview

The skill is mostly a coherent video-ad helper, but its install instructions ask users to run mutable remote code directly in a shell.

Review this skill before installing. Its video-ad guidance and infsh-only runtime scope are coherent, but avoid running the curl-to-shell Quick Start unless you independently trust and inspect the installer. Prefer a pinned, checksum-verified manual install of the CLI, and review any related skills before adding them with npx.

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
Unverified Remote Installation Script Executed Directly by a Shell<![CDATA[ ## Vulnerability Details **File Location**: `SKILL.md:14` **Vulnerability Type**: Remote payload retrieval and execution **Risk Level**: High ### Vulnerable Code ```bash curl -fsSL https://cli.inference.sh | sh && infsh login ``` ### Technical Analysis The command pipes a mutable HTTP response directly into `sh`. Although the document states that the installer verifies the subsequently downloaded binary, the bootstrap script itself is executed before its content, checksum, signature, or version is independently verified. HTTPS protects the connection in transit but does not eliminate risks from a compromised hosting server, account or DNS takeover, certificate misuse, or malicious changes to the remotely hosted script. Consequently, the effective code executed by this reviewed Skill can change at any time without changing `SKILL.md`. This installation behavior also exceeds the minimum privileges required to provide the Skill's declared video-ad specifications. The installer does not require elevated privileges according to the document, but it still receives all permissions available to the Agent's current operating-system user. ### Attack Path 1. A user or Agent loads the Skill and follows its Quick Start instructions. 2. `curl` retrieves the current content served by `https://cli.inference.sh`. 3. The response is passed directly to `sh` without inspection or pre-execution integrity verification. 4. If the endpoint or its delivery infrastructure is compromised, attacker-controlled shell commands execute as the current Agent user. 5. Those commands can access files and credentials available to that user, modify the environment, install additional software, or launch other processes. 6. The command then starts `infsh login`, potentially exposing the subsequent authentication workflow to a compromised installation. ### Impact Assessment Successful exploitation provides arbitrary command execution with the privileges of the user running the co ...[truncated 293 chars]
Remediation
<![CDATA[ ## Remediation Suggestions 1. Remove the `curl | sh` installation method from the primary workflow. 2. Pin the CLI to a specific, reviewed version and immutable release URL. 3. Download the artifact without executing it: ```bash curl --fail --location --output infsh "<version-pinned-release-url>" ``` 4. Publish and pin the expected SHA-256 digest in the reviewed Skill rather than obtaining both the artifact and expected digest from the same mutable source. 5. Prefer signed releases and verify the signature with a trusted public key distributed through an independent channel. 6. Install the verified artifact using only user-level permissions. 7. Keep `infsh login` as a separate command and require explicit user approval before beginning authentication. 8. Provide manual inspection and installation steps as the default; if an installer remains available, treat it only as an optional alternative accompanied by a clear warning. ]]>

T08 · Insecure Dependencies

Warning
Location
SKILL.md:247
Finding
Unpinned Third-Party Skill Installation Through npx<![CDATA[ ## Vulnerability Details **File Location**: `SKILL.md:247-250` **Vulnerability Type**: Insecure dependencies and mutable supply-chain inputs **Risk Level**: Medium ### Vulnerable Code ```bash npx skills add inference-sh/skills@ai-video-generation npx skills add inference-sh/skills@video-prompting-guide npx skills add inference-sh/skills@text-to-speech npx skills add inference-sh/skills@prompt-engineering ``` ### Technical Analysis These optional commands invoke the `skills` package through `npx` without pinning that executable package to an exact reviewed version. They also identify external Skill content by names rather than immutable, reviewed revisions or commit hashes. This creates multiple mutable supply-chain boundaries. A future package release or change to the referenced external Skill repository could introduce content that was not present during this audit. Depending on the behavior of the `skills` installer, fetched content may subsequently influence Agent instructions or introduce additional executable resources. The audit did not establish that the named dependencies are currently malicious. The vulnerability is the absence of immutable version selection and integrity validation before third-party content is fetched and installed. ### Attack Path 1. A user follows one of the optional Related Skills commands. 2. `npx` resolves and executes the currently available version of the `skills` package rather than a version reviewed with this project. 3. The installer retrieves the named external Skill from a mutable upstream source. 4. If the package registry account, installer package, source repository, or publishing credentials are compromised, modified content is delivered to the user. 5. The imported Skill can then affect later Agent sessions and may exercise whatever tools or permissions its own configuration permits. ### Impact Assessment The immediate scope includes the user's Skill installation environment and future Agent beh ...[truncated 426 chars]
Remediation
<![CDATA[ ## Remediation Suggestions 1. Pin the `skills` package invoked through `npx` to an exact reviewed version. 2. Pin each external Skill to an immutable release, commit hash, or content digest. 3. Use lockfiles or an equivalent integrity-controlled dependency manifest where supported. 4. Verify downloaded content against independently published checksums or cryptographic signatures. 5. Review each third-party Skill's instructions, scripts, and allowed tools before installation. 6. Require explicit user confirmation before fetching or enabling external Skills. 7. Avoid automatically granting imported Skills shell access, credentials, or broad filesystem permissions. 8. Periodically re-audit dependencies before intentionally updating pinned versions. ]]>
Vulnerability Patterns
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (7)

External Script Fetching

High
Category
Supply Chain
Content
## Quick Start

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

# Generate a vertical video ad scene
infsh app run bytedance/seedance-1-5-pro --input '{
Confidence
97% confidence
Finding
`curl -fsSL https://cli.inference.sh | sh` fetches executable code from a remote server and immediately runs it in the user's shell, creating direct remote code execution risk. If the hosting service, TLS trust chain, DNS, or distribution pipeline is compromised, the attacker gains code execution on systems following the skill.

Vague Triggers

Medium
Confidence
97% confidence
Finding
The manifest description includes a very broad trigger list with generic phrases like `video ad`, `paid media`, and `video marketing`, increasing the chance this skill auto-activates in unrelated contexts. Over-broad activation can route users into tool-using workflows unexpectedly, raising the chance of unintended command suggestions or execution paths.

Missing User Warnings

Medium
Confidence
93% confidence
Finding
The skill explicitly instructs users to pipe a remote script from `https://cli.inference.sh` directly into `sh`, which is a classic unsafe installation pattern. Even with adjacent reassurance text, users are being told to execute downloaded code before independently verifying its contents or integrity.

Rp1

Medium
Category
MCP Rug Pull
Confidence
86% confidence
Finding
The skill recommends running `npx skills add ...` without pinning a specific version or commit, which causes users to fetch and execute whatever package/version is current at runtime. If the package, dependency chain, or referenced skill source is compromised, users could install or execute unreviewed code or content.

Rp1

Medium
Category
MCP Rug Pull
Confidence
86% confidence
Finding
This unpinned `npx skills add ...` invocation introduces supply-chain risk because the resolved package and retrieved skill content may change over time. A later malicious or vulnerable update could be pulled automatically by anyone following the instructions.

Rp1

Medium
Category
MCP Rug Pull
Confidence
86% confidence
Finding
Running `npx` against an unpinned package means arbitrary newer code can be executed in the user's environment without prior review. In a skill distribution context, this is a real supply-chain weakness even if the current package is legitimate.

Rp1

Medium
Category
MCP Rug Pull
Confidence
86% confidence
Finding
Because this command relies on an unversioned remote package and skill reference, the behavior is non-deterministic and could be altered by upstream compromise. Users may unknowingly execute changed code or install altered skills in future runs.

Static analysis

No suspicious patterns detected.