Security audit
Security checks for vulnerabilities and agentic risk
This is a local work-report generator with disclosed report-history reuse and local storage, though users should notice its broad trigger wording and automatic history reads.
Install this only if you are comfortable with a report helper reading prior reports from .workbuddy/reports/ to build summaries. Review drafts before approving saves, and avoid putting highly sensitive business or personnel details in stored reports unless that reuse is intended.
The trigger definition is overly broad and includes language like arbitrary related input, which can cause the skill to activate on normal conversation unintentionally. In this skill's context, unintended activation is risky because activation can lead to automatic reading of local historical reports and generation based on prior sensitive work content the user did not freshly request.
The trigger table allows activation from broad natural-language rambling and unconstrained keywords, which makes invocation rules vague and prone to false positives. In combination with the skill's automatic history-reading behavior, this increases the chance of surfacing prior sensitive work reports in contexts where the user only mentioned related phrases casually.
The file mandates a specific colloquial Chinese output style as a required rule, but does not present a user language or locale choice in the active behavior. A forced language/style constraint can violate language/locale policy when users are not given an opt-in or alternative.
Defaulting ambiguous keyword input to a daily report creates unclear activation boundaries and can cause the skill to act when the user did not intend to generate or persist a report. Because this skill also stores and reuses report content, accidental invocation can expose or memorialize sensitive work details unnecessarily.
The skill instructs the agent to automatically read and reuse prior reports from local storage without a fresh disclosure or confirmation step. Those reports can contain sensitive business context, incidents, vulnerabilities, timelines, or personnel details, so automatic reuse may reveal prior private data in a new response even when the user only provided minimal new input.
L163-L170 instructs the skill to read historical reports automatically and '静默进行', while L327 explicitly says that if historical reading fails, the user must be informed and the skill must not silently skip it. These instructions describe opposite intended behaviors for the same situation, creating intent ambiguity in the skill specification.
No suspicious patterns detected.
