The Agent Skills Directory

[DATA_EXPOSURE_AND_EXFILTRATION]: The setup.sh script stores authentication tokens in the /tmp/ directory (e.g., .tdoc_auth_token). These files are world-readable on many systems, allowing other local users or processes to harvest the user's credentials.
[DYNAMIC_EXECUTION]: The OperationSheet tool in sheet/api/operation-api.md executes JavaScript code (js_script) that is generated at runtime by the agent. This allows for arbitrary logic execution within the spreadsheet environment based on user input.
[INDIRECT_PROMPT_INJECTION]: The scrape_url tool in references/workflows.md fetches content from arbitrary external websites and saves it into the document context. This content is subsequently read by the agent through smartcanvas.read, creating an ingestion surface for untrusted instructions without explicit sanitization or boundary markers.
[UNVERIFIABLE_DEPENDENCIES_AND_REMOTE_CODE_EXECUTION]: The setup.sh script performs a global installation of the mcporter package using npm install -g at runtime without specifying a version or verifying the source.

tencent-docs