Security audit
Security checks for vulnerabilities and agentic risk
This is a planning helper, but it can automatically write project and cross-project memories that may affect future sessions without asking at the time.
Install only if you want this skill to shape general problem-solving behavior and are comfortable with it saving selected conclusions, project conventions, and cross-project preferences into `.workbuddy` memory files. Review those files periodically and avoid using it on sensitive tasks unless the agent asks before persisting anything.
The trigger phrases are broad, common requests that can cause this skill to activate in many ordinary conversations. Because the skill then changes agent behavior, including possible memory writes and external lookups, overbroad activation increases the chance of unintended execution without clear user intent.
The activation guidance and trigger examples are specified in Chinese, including explicit phrases the user must say to invoke the skill. The file does not indicate that language choice is optional or offer equivalent user-selectable locale behavior, which can be a locale policy issue if the skill is expected to work without forcing a specific language.
The skill description explicitly instructs the agent to distill key points into cross-session memory. Even though it tries to limit what is stored, persistent retention of user/task-derived information in plain language can capture sensitive operational context and create privacy or data-minimization issues over time.
This section operationalizes persistent logging of project conclusions, changes, preferences, and long-term conventions into specific memory files. Because these are durable writes that can influence later sessions and may include sensitive project information, they create a meaningful privacy and integrity risk if performed automatically or without clear consent.
The skill instructs the agent to write to project and home-directory memory files, including persistent locations, without requiring a clear warning or explicit consent at the time of modification. Persistent writes can alter future agent behavior, leak sensitive task details into storage, and create hard-to-notice cross-session side effects.
The boundary section says the skill is '纯本地、零依赖，不检索内容库也不回写', implying a local-only, dependency-free method. But earlier sections require escalating to external official documentation, T1 communities, and search engines when needed, which is not purely local behavior and does rely on external information sources.
Line L09 states that Chinese scenarios should prioritize Chinese sources, which is a natural-language locale preference applied by default. The file does not indicate that users can choose another language or opt out of this behavior.
No suspicious patterns detected.
