The Agent Skills Directory

[DATA_EXFILTRATION] (MEDIUM): The skill archives and transmits the contents of the current working directory to vibelink.to using curl. While it attempts to exclude .env files, any other sensitive data or hardcoded credentials within the project files are exfiltrated to a non-whitelisted external domain.
[COMMAND_EXECUTION] (MEDIUM): The skill executes arbitrary commands found in package.json scripts to start a development server and capture screenshots. A malicious repository could contain harmful shell commands in the dev or start scripts which would be executed with the user's privileges.
[CREDENTIALS_UNSAFE] (LOW): The skill manages and stores persistent 'author tokens' in plain text within the ~/.vibelink-tokens/ directory. While used for legitimate service authentication, these are locally stored secrets managed by the skill.
[PROMPT_INJECTION] (LOW): Indirect Prompt Injection Surface: The skill ingests untrusted content from package.json and README.md (Ingestion Points) and uses this data to influence agent behavior without boundary markers or sanitization. This is particularly risky given the skill's capabilities (Capability Inventory) to execute shell commands and perform network uploads.

vibelink-push