Back to skill

Security audit

GoAI Video Gen

Security checks for vulnerabilities and agentic risk

Overview

This skill mostly does what it says, but it saves downloaded videos using a server-provided ID in a way that could overwrite unexpected local files if the API endpoint is malicious or compromised.

Review before installing. Use the default GoAI endpoint unless you intentionally trust another GOAI_BASE_URL, avoid passing sensitive local images unless you are comfortable uploading them to GoAI, and run the skill from a workspace where generated media files can be safely written. The publisher should validate task_id values and confine downloads to a dedicated output directory before this is treated as routine-risk.

Vulnerability Patterns
  • 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
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
Findings (1)

T09 · Insecure Skill Coding Practices

Error
Location
scripts/generate_video.py:78
Finding

Server-Controlled Task ID Enables Path Traversal and Arbitrary File Overwrite

Content
View full analysis
str: if not media_url: raise GoAIError("download URL is required") output_path = Path(output_name).expanduser() if not output_path.is_absolute(): output_path = (Path.cwd() / output_path).resolve() output_path.parent.mkdir(parents=True, exist_ok=True) try: response = self._http.get(media_url) except httpx.RequestError: raise GoAIError(f"download failed: {media_url}") from None if response.status_code not in {200, 201}: raise GoAIError(f"download failed: {media_url}") output_path.write_bytes(response.content) return str(output_path) ``` ### Technical Analysis The `task_id` value originates in the remote API response and is accepted after only checking that it is nonempty. It is then interpolated directly into the local output filename. An attacker-controlled value can contain path separators and parent-directory components. For example, a task identifier shaped like `x/../../../target` produces an output ...[truncated 2162 chars]
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
  • Rogue AgentSelf-Modification, Session Persistence
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
Findings (12)

Tp4

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

The declared purpose is video generation, but the behavior also includes runtime bootstrapping, dependency preparation, and self-healing execution logic. This mismatch can mislead operators and reviewers about what the skill will actually do on the host, especially because bootstrapping can alter local runtime state and execute additional tooling.

Content

No source excerpt is available for this finding.

os.system() or os exec-family call

High
Category
Dangerous Code Execution
Confidence
85% confidence
Finding

os.system() and os exec-family calls run shell commands with the process's full privileges, enabling arbitrary command execution.

Content

Scanner excerpt · scripts/bootstrap.py (reported line 50)May include surrounding context.

python
print(_uv_install_hint(), file=sys.stderr)
        raise SystemExit(1)

    os.execvpe(
        uv_bin,
        [
            uv_bin,

Env Variable Harvesting

High
Category
Data Exfiltration
Confidence
60% confidence
Finding

Code enumerates, copies, or searches environment variables for secrets. Bulk environment access can collect credentials unrelated to the skill's stated purpose.

Content

Scanner excerpt · scripts/bootstrap.py (reported line 61)May include surrounding context.

python
str(script_path),
            *sys.argv[1:],
        ],
        os.environ.copy(),
    )

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
92% confidence
Finding

The skill declares capabilities that clearly require shell, network, environment, and file access, but it does not constrain or declare an explicit tool scope. That increases the blast radius if the skill is invoked unexpectedly or modified later, because a consumer cannot readily enforce least privilege from the manifest.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The manifest description says to use the skill when the user asks to create, generate, render, or make videos, animations, clips, shorts, or motion content, and includes broad Chinese phrases like 做视频. These triggers are not tightly scoped and could match ordinary conversation without clear exclusion conditions or negative examples in the invocation description.

Content

No source excerpt is available for this finding.

Session Persistence

Medium
Category
Rogue Agent
Confidence
60% confidence
Finding

Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Content

Scanner excerpt · SKILL.md (reported line 15)May include surrounding context.

md
Pass the user's video prompt through as-is unless the user explicitly asks for prompt optimization, prompt polishing, rewriting, translation, or additional creative detail. Do not silently embellish short prompts.

Do not translate the user's prompt to another language. If the user writes in Chinese, pass Chinese to `--prompt`. If the user writes in English, pass English to `--prompt`. You may strip only lightweight request wrappers such as `帮我`, `给我`, `请`, `生成`, or `做个`, but do not rewrite the core prompt content.

Examples:

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The skill states that local image paths are uploaded to a remote service, but the description and user-facing guidance do not prominently warn about this data egress. Users may unintentionally submit sensitive local images or proprietary content to a third-party API without informed consent.

Content

No source excerpt is available for this finding.

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
80% confidence
Finding

Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.

Content

Scanner excerpt · SKILL.md (reported line 60)May include surrounding context.

md
- macOS: `brew install uv`
- Windows: `winget install astral-sh.uv`

After `uv` is installed, rerun the same command. Do not ask the user to install Python packages by hand; `uv` is responsible for preparing Python and the skill dependencies.

If execution reports a missing Python module such as `httpx`, do not run `pip install`, `pip install --user`, or `pip install --break-system-packages`. Re-run the skill through `uv`; the dependency set comes only from `pyproject.toml` and `uv.lock`.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The code automatically treats any non-URL image input as a local file and uploads it to a remote GoAI-controlled endpoint, but provides no user-facing disclosure or confirmation at the point of exfiltration. In an agent skill context, users may supply local paths expecting local processing, so silent upload of filesystem content can leak sensitive images or unexpectedly transmit private data off-device.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
95% confidence
Finding

The file explicitly instructs the skill not to translate the user's prompt to another language and to preserve Chinese or English input unless the user asks for translation. This does not force a specific language or locale; instead, it respects the user's chosen language.

Content

No source excerpt is available for this finding.

Unverifiable Dependency: httpx has 2 known advisory(ies) (CVE-2021-41945 (Improper Input Validation in httpx); CVE-2021-41945 (Encode OSS httpx <=1.0.0.beta0 is affected by improper input validation in `http)), but the manifest does not pin a version, so it is unknown whether the installed release is affected

Low
Category
Supply Chain
Confidence
40% confidence
Finding

Dependency has known vulnerabilities (CVEs). Using packages with unpatched security flaws exposes the environment to known exploits.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
84% confidence
Finding

The code downloads remote media from a URL and writes it to a caller-controlled filesystem path without any user-facing notice. While this is expected functionality for a video generation skill, silent writes can still surprise users, overwrite files in unsafe locations if higher layers do not constrain output paths, and persist untrusted remote content locally.

Content

No source excerpt is available for this finding.