Security audit
Security checks for vulnerabilities and agentic risk
This is mainly an AI cost-estimation skill, but its requested tool access and permission/risk metadata are inconsistent enough that users should review it before installing.
Install only if you specifically want an LLM cost-estimation and cost-optimization helper. Review or narrow the allowed tools before use, especially Bash and Edit, and require explicit confirmation before making local deployment, caching, or file-changing recommendations. Do not provide secrets or billing credentials to this skill; the inspected script does not need them.
The file title and all user-facing instructions are presented solely in Chinese, and there is no indication that language selection is optional or region-specific. Under the stated policy, forcing a specific language without user opt-in can be a natural-language policy violation.
The trigger list is very broad and includes common phrases like cost reduction, local models, batching, and API bills, which can match many ordinary user requests. This increases the chance of unintended activation, causing the agent to apply this skill in contexts where it is not appropriate and potentially steer decisions or modifications without clear user intent.
The trigger list is broad and includes generic phrases such as cost, local model, batching, and offline AI that may appear in unrelated conversations. Overbroad activation can cause the agent to invoke the wrong skill, leading to unintended workflow changes, misleading recommendations, or unnecessary execution of attached tooling in contexts the user did not intend.
At L030, the documentation states the skill is a 'batch patch' default-medium item involving '调用 API/读写文件/本地执行'. That contradicts the manifest and documented interface elsewhere in this file, which define a local CLI cost estimator with network: false, filesystem: false, and subprocess: false and no evidence of those side-effecting capabilities.
The activation flow says to detect keywords and confirm optimization intent, but it does not define disqualifying conditions or precise scope limits. In agent environments, ambiguous routing logic can misfire on incidental keywords, causing context-inappropriate activation and potentially influencing user decisions with irrelevant cost-cutting guidance.
The file content is entirely presented in Chinese and includes user-facing invocation guidance without indicating that other languages are supported or that Chinese is a required locale. This can amount to a language-policy constraint without opt-in or documented justification.
The trigger text is very broad, covering many common cost-optimization phrases and adjacent concepts such as local models, batching, routing, and offline AI. In an agentic system, this can cause the skill to activate in situations where the user did not actually request this specific workflow, leading to inappropriate tool use, bad recommendations, or unsafe workflow substitution.
The file's docstring and all argparse help text are written entirely in Chinese, which imposes a locale/language choice on users without opt-in. Under the policy criteria, forced language selection is a natural-language policy violation unless the locale constraint is documented and justified or the user is offered a choice.
The file presents the skill description in both English and Chinese and brands the project as China-associated, but it does not state that language use is optional or explain a region-specific requirement. Under the stated policy, language or locale constraints should either be optional for users or clearly justified.
