T09 · Insecure Skill Coding Practices
- Location
scripts/videogen.py:91- Finding
Arbitrary API Base URL Allows Bearer Credential Exfiltration
- Content
View full analysis
- Remediation
View remediation
Security audit
Security checks for vulnerabilities and agentic risk
The skill is mostly a disclosed AI-HIVE video-generation tool, but it can implicitly submit external jobs and can send the bearer API key to a user-configurable API base URL.
Install only if you intend to use AI-HIVE for video generation and are comfortable uploading prompts and chosen media to that service. Before running it, avoid implicit or vague invocations, confirm any task that may cost money, keep the API key private, do not set a custom base URL unless you fully trust it, and prefer installing dependencies in an isolated environment with a pinned `requests` version.
scripts/videogen.py:91Arbitrary API Base URL Allows Bearer Credential Exfiltration
SKILL.md:200Unpinned Third-Party Dependency Installation
The skill documents and encourages capabilities that imply network access, shell execution, file reads/writes, and environment/config handling, but it does not declare any explicit tool scope or permission boundaries. In an agent environment, this creates a confused-deputy risk: a broadly invocable skill could perform sensitive local and network actions without clear least-privilege constraints or user-visible consent boundaries.
The description is extremely broad and overlaps with generic content creation, advertising, ecommerce, AI model, and media-production requests, making over-invocation likely. In an agent setting, broad triggering is dangerous because this skill can upload files, persist API keys/configuration, make external network requests, poll tasks, and download outputs, so accidental invocation can cause data exfiltration, unwanted charges, or unauthorized processing.
The metadata sets language: zh-CN, and the skill content is primarily written to operate in Chinese, but the document does not indicate that users can opt into another language or that the restriction is required for a region-specific compliance reason. This can violate language/locale policy when a skill effectively enforces a locale by default.
The search coverage section enumerates a very broad set of keywords spanning many platforms, tools, models, and commerce contexts without guardrails, which can cause the routing system to match unrelated requests. Because the skill performs real external actions and handles local media plus credentials, over-broad keyword capture materially increases the chance of unintended uploads, downloads, API usage, and billing.
The manifest explicitly advertises automatic submission, polling, and result download, but provides no user-facing warning that content may be uploaded to an external service and that persistent identifiers/results may be stored. In this skill's context, those behaviors can expose user media, prompts, and generated outputs to unintended external processing or incur costs without sufficiently informed consent.
The default prompt is broad and action-oriented: it tells the agent to complete a video task, save a taskId, and retrieve results, without requiring an explicit confirmation step. Combined with implicit invocation, this can cause the skill to activate in response to loosely related user requests and trigger external job submission or data handling the user did not clearly authorize.
The file-level docstring is entirely in Chinese and presents the tool as a general-purpose skill, while user-facing prompts and help text throughout the CLI are also Chinese-only. For a general skill, forcing a specific language without opt-in is a natural-language locale policy issue under the stated rule.
The placeholder value for api_key includes Chinese text (请替换为你的APIKey), which imposes a specific language in user-facing configuration guidance. This can violate language/locale policy when no user choice or documented locale constraint is provided.