Security audit
Security checks for vulnerabilities and agentic risk
This skill is a clearly scoped OpenClaw context-window configuration helper that reads limited model settings and only writes after user confirmation.
Before installing, understand that the skill can modify OpenClaw model context-window configuration after you confirm. Review the decision card and diff carefully, especially because the docs include broad phrases like context check or context optimization that may start a read-only scan.
This manifest/markdown-style progress file records recommended test phrases such as “上下文优化” and “比例 50%”, which are relatively generic and could overlap with ordinary discussion rather than an explicit skill invocation. The line does not include limiting conditions or negative examples here, so the trigger scope is ambiguous in this file.
The README explicitly says plain-language trigger phrases are enough and lists broad variants like '检查上下文' / 'optimize context', which can match ordinary conversation rather than the tightly scoped intents claimed in the metadata. That mismatch increases the chance the skill activates unexpectedly and proceeds into a workflow that reads live configuration and external vendor sources, creating unintended config-modification prompts and possible user confusion or social engineering opportunities.
The README advertises broad natural-language triggers such as “检查上下文 / 优化上下文 / 上下文体检”, while the skill metadata simultaneously says it must not activate on generic discussion of context, memory, prompts, or token usage. This mismatch creates a real risk of accidental invocation during ordinary conversation, which could lead the agent into reading configuration and making network lookups unexpectedly, and potentially progressing toward a config patch if the user responds ambiguously.
The design document is written entirely in Chinese and specifies a Chinese trigger phrase set, but it does not indicate that users may choose their preferred language or locale. Under the policy, forcing a specific language without opt-in is a natural-language policy concern unless the locale restriction is explicitly documented and justified.
The design doc expands activation to broad phrase matching such as any occurrence of '上下文' plus '检查/体检/查/看看', which conflicts with the manifest's narrower explicit-intent requirement. This can cause the skill to activate on ambiguous requests and initiate network reads, configuration inspection, or a write-approval workflow the user did not specifically ask for, increasing the risk of unintended sensitive operations.
The note about “hero.svg 中文-only 的本地化提示” suggests the skill assets were recognized as Chinese-only, but this file does not indicate user opt-in or a justified locale restriction. That can conflict with language/locale policy expectations when a skill is not explicitly region-specific.
This SVG includes prominent user-facing text only in Chinese ("把上下文窗口配成官方标称的 60%" and "检测 → 决策 → 执行") without any indication that the skill is China-specific or that users can choose their language. That is a natural-language locale policy concern because it imposes a specific language presentation without documented opt-in or justification.
