Security audit
Security checks for vulnerabilities and agentic risk
This is a transparent document co-authoring workflow with some normal privacy considerations around using connected tools and sub-agents.
Install only if you are comfortable with a writing assistant that may ask to read linked documents or team discussions and may test drafts with a fresh agent. For confidential material, specify exactly which sources it may use and redact secrets, credentials, personal data, or sensitive strategy before reader testing.
The description says to trigger when the user wants to write documentation or mentions 'similar documentation tasks,' which is open-ended and not tightly bounded. This broad phrasing can overlap with ordinary writing assistance requests and may cause unintended invocation without clear exclusion criteria or negative examples.
Phrases like 'write up' are common everyday language, and 'User seems to be starting a substantial writing task' is a subjective condition with no clear boundary. Without tighter constraints or examples of what should not trigger the skill, the workflow may activate inappropriately.
The skill instructs the agent to pull context from external integrations such as messaging platforms and document stores, which can expand access to sensitive organizational data beyond what is necessary for the immediate drafting task. Even with user prompting, this broad retrieval pattern increases the chance of over-collection, inadvertent inclusion of confidential material, or cross-source data exposure during document creation.
The instruction to search connected tools for unknown entities or projects is broader than necessary for co-authoring and can trigger exploratory access into unrelated organizational systems. Even though the skill says to wait for confirmation, it still normalizes wide-scope discovery behavior that may surface unnecessary confidential information or create privacy issues.
The skill directs autonomous testing with fresh sub-agents using the full document content, which can replicate sensitive material into additional agent contexts without a clear need-to-know boundary. This increases exposure surface and may cause confidential content to be processed by multiple agent instances, especially if the document contains internal strategy, security details, or personal data.
The manifest focuses on guiding users through drafting and refining content, but the instructions explicitly direct creation of artifacts or markdown files and repeated in-place edits. While related to authoring, persistent file creation/modification is an operational behavior not stated in the manifest description.
No suspicious patterns detected.
