T09 · Insecure Skill Coding Practices
- Location
scripts/imagegen.py:91- Finding
AI-HIVE bearer credential disclosure through an unrestricted API base URL in the image client
- Content
View full analysis
- Remediation
View remediation
Security audit
Security checks for vulnerabilities and agentic risk
The skill mostly matches its AI-HIVE marketing-generation purpose, but it needs Review because it can send API credentials to a user-configurable API host and automatically downloads unvalidated remote results.
Review this skill carefully before installing. Use only the official AI-HIVE endpoint, avoid setting --base-url or AI_HIVE_BASE_URL unless you fully trust that server, keep the API key limited and rotatable, use --no-download or a dedicated output directory for untrusted tasks, and upload only media you have rights to use.
scripts/imagegen.py:91AI-HIVE bearer credential disclosure through an unrestricted API base URL in the image client
scripts/videogen.py:91AI-HIVE bearer credential disclosure through an unrestricted API base URL in the video client
scripts/imagegen.py:309Unvalidated task-result URLs and path components in image-result downloads
scripts/videogen.py:309Unvalidated task-result URLs and path components in video-result downloads
scripts/imagegen.py:7Unpinned third-party dependency installation guidance in the image client
scripts/videogen.py:7Unpinned third-party dependency installation guidance in the video client
The reported presence of undeclared general chat, multimodal input, and user/billing queries broadens the attack surface beyond the advertised domain-specific marketing use case. When a skill claims copyright and authenticity guardrails but does not actually enforce them, reviewers and users may falsely assume safety controls exist, enabling misuse of uploaded media, account data, or paid operations.
The reported presence of undeclared general chat, multimodal input, and user/billing queries broadens the attack surface beyond the advertised domain-specific marketing use case. When a skill claims copyright and authenticity guardrails but does not actually enforce them, reviewers and users may falsely assume safety controls exist, enabling misuse of uploaded media, account data, or paid operations.
The reported presence of undeclared general chat, multimodal input, and user/billing queries broadens the attack surface beyond the advertised domain-specific marketing use case. When a skill claims copyright and authenticity guardrails but does not actually enforce them, reviewers and users may falsely assume safety controls exist, enabling misuse of uploaded media, account data, or paid operations.
The reported presence of undeclared general chat, multimodal input, and user/billing queries broadens the attack surface beyond the advertised domain-specific marketing use case. When a skill claims copyright and authenticity guardrails but does not actually enforce them, reviewers and users may falsely assume safety controls exist, enabling misuse of uploaded media, account data, or paid operations.
The skill advertises executable commands and implies use of environment variables, file I/O, network access, and shell execution, but it does not declare any explicit tool scope or permission boundaries. In an agent setting, missing least-privilege declarations can cause the runtime to grant broader capabilities than necessary, increasing the blast radius if the skill is misused or if downstream scripts behave unexpectedly.
An overly broad trigger description can cause the skill to activate for many loosely related user requests, including ones not intended for this workflow. In agent systems, overbroad activation increases the chance that powerful tools, external links, or paid-generation flows are invoked in the wrong context, exposing data or causing unintended actions.
The usage scenarios enumerate wide activation conditions across many generic marketing and content terms without strong limiting conditions. Because this skill also references external services, scripts, and media handling, broad matching raises the likelihood of accidental invocation and inappropriate processing of user content or credentials in unrelated conversations.
The default prompt hardcodes Chinese output without checking the user's language preference or obtaining opt-in. While not a direct code-execution issue, it can override user intent, reduce transparency, and cause miscommunication in a workflow involving marketing plans, content strategy, and potentially contractual or rights-sensitive deliverables.
The skill enables implicit invocation with no documented trigger constraints, exclusions, or narrowing examples. That can cause the agent to activate this marketing skill on loosely related user requests, increasing the chance of unintended tool behavior, surprise execution paths, and inappropriate handling of user content or references in a domain that involves asset uploads, model routing, and task tracking.
subprocess module calls execute external commands. Without careful input validation, this enables command injection.
def run(args: list[str]) -> None:
print(" ".join(args))
subprocess.run(args, check=True)
def main() -> None:
The module docstring and user-facing instructions are written entirely in Chinese, and later prompts/messages throughout the skill also assume Chinese-only interaction. For a general-purpose skill, this creates a language/locale policy concern because the skill does not offer opt-in or alternative language selection.
The skill hard-codes core user-facing content in Chinese via the INDUSTRY, MODE, FOCUS, CHANNELS, and BOUNDARY constants, with no option for the user to choose language or locale. This creates a natural-language policy issue because it effectively forces a specific language without opt-in or documented justification.
The module docstring and all user-facing operational text are written only in Chinese, with no indication that another language is available. Under the stated policy, forcing a specific language without offering user choice is a natural-language policy violation.
The skill automatically downloads files from server-provided resultUrl values into a local output directory after task completion unless --no-download is set. In an agent or automation context, this can cause unintended local writes and retrieval of untrusted remote content without an explicit confirmation step, increasing risk of disk abuse, unsafe artifact handling, or downstream opening of malicious media files.
This markdown file presents all guidance in Chinese and does not indicate that the user can choose another language. That can violate a language/locale policy when skills are expected to avoid forcing a specific language without user opt-in.