The Agent Skills Directory

[SAFE]: The skill implements a legitimate notification utility. It stores sensitive information (Telegram tokens and Discord webhooks) in the user's home directory (~/.config/claude-notify/config.json), which is a standard practice for local configuration.
[DATA_EXPOSURE_&_EXFILTRATION]: While the skill communicates with external APIs (api.telegram.org and Discord webhooks), it only sends notification data (event status, project paths, and manual messages). There is no evidence of unauthorized exfiltration of sensitive local files or credentials. The skill's instructions specifically advise hiding sensitive parts of the tokens when displaying configuration.
[COMMAND_EXECUTION]: The skill uses Node.js scripts to handle logic. These scripts are invoked with static arguments (--test, --manual) and do not process unsanitized shell commands or user input in a way that leads to injection.
[INDIRECT_PROMPT_INJECTION]: The skill processes hook data from Claude Code. While this is an ingestion surface, the logic is limited to extracting specific fields (status, error messages) and formatting them for notification, posing no risk to the agent's behavior.
[EXTERNAL_DOWNLOADS]: The skill mentions an external plugin repository for automatic notifications, but these are provided as optional manual installation instructions for the user via the official claude plugin command, following standard extension patterns.

custom-skills-notify