Security audit
Security checks for vulnerabilities and agentic risk
This skill is a disclosed learning-coach workflow that reads user-provided materials with read-only access and creates lesson files, with no evidence of hidden execution, exfiltration, or system persistence.
Install this if you want a structured learning workflow that can read materials you provide and create lesson files. Before using it with private Feishu or local documents, confirm you are comfortable granting read-only access and having summarized lesson artifacts stored in the workspace.
The skill description uses very broad activation cues covering generic learning, research, business analysis, and training requests. In agent environments that auto-route by description, this can cause the skill to trigger for many ordinary user requests and then perform file creation, source gathering, and internal-document handling that the user may not have specifically intended.
Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.
- Base or bitable material: read table fields and records with read-only base access.
- File, image, or attachment: try download, OCR, or parsing before treating it as OAuth.
If authorization is needed, ask through the tool's normal minimum read-only flow. Do not ask the user to choose technical permission options, do not request write/admin scopes for learning, and batch the minimum read-only permissions when the tool supports it.
User-facing authorization message should stay short:
Both the short description and default prompt are written only in Chinese, and the prompt directs the interaction flow in Chinese without indicating that other languages are supported. This can constitute a language-policy issue because the skill appears to enforce a specific language without user opt-in or documented justification.
The default prompt contains a broad activation phrase and directly instructs the agent to invoke the skill without clear gating conditions, which can cause over-triggering or unintended routing. In a skill that performs deep research, lesson generation, and source-grounded output, this increases the chance of the agent entering a high-authority workflow when the user did not explicitly request it, potentially amplifying prompt injection from supplied materials or causing unnecessary data access.
The skill instructs the agent to create and update workspace files such as lesson HTML, LEARNING_STATUS.md, and LEARNING_SOURCES.md without requiring an upfront warning that user files will be modified. This creates a consent and integrity risk because the agent may alter the workspace, leave persistent artifacts, or overwrite prior learning records when the user only asked for conversational help.
No suspicious patterns detected.