Security audit
Security checks for vulnerabilities and agentic risk
This is a low-impact planning helper with overly broad triggers and unrelated pasted learning content, but no executable code, credential access, persistence, or hidden behavior.
Before installing, consider narrowing the activation phrase and removing the unrelated Bilibili sections so the skill only loads when you explicitly want an implementation plan. The reviewed artifact does not show sensitive access or automatic system changes.
The trigger phrase 'write plan' is broad enough to collide with ordinary user requests, making the skill likely to activate outside narrowly intended contexts. Over-broad activation can cause unauthorized skill invocation, unexpected behavior selection, or prompt hijacking opportunities when normal conversation is misclassified as a skill trigger.
The invocation examples include generic phrases like '规划', 'plan', and '步骤', which are common in everyday conversation and highly likely to overlap with unrelated requests. This ambiguity increases accidental triggering risk and makes it easier for other content in a conversation to steer the agent into executing this skill unexpectedly.
The skill file contains large amounts of unrelated Bilibili 'learning' content that does not support the declared implementation-plan functionality. In an agent skill, irrelevant embedded content increases prompt-surface area, can confuse routing or downstream reasoning, and creates a channel for accidental or adversarial instruction contamination even if this specific text is not overtly malicious.
The skill defines trigger scenarios in Chinese while also including the English term "plan", but it does not state whether the skill is intended for a specific language audience or how user language preference is handled. This can create an implicit language/locale behavior without clear opt-in or justification.
No suspicious patterns detected.
