The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides a feature to configure and execute local shell commands for event notifications.
Evidence: The "Generic CLI Command Flow" allows users to specify an executable and a list of arguments for runtime execution.
Sanitization: Includes validation rules (no spaces, no shell metacharacters) to mitigate command injection risks.
[EXTERNAL_DOWNLOADS]: Interacts with remote APIs to test notification connectivity.
Evidence: Uses curl to send test messages to api.telegram.org, discord.com, and hooks.slack.com.
Context: These are well-known services, and the interactions are restricted to connectivity testing using user-provided credentials.
[DATA_EXFILTRATION]: Facilitates the collection and storage of sensitive authentication data.
Evidence: Instructs users to paste Telegram bot tokens, Discord webhook URLs, and Slack webhook URLs, which are saved to ~/.claude/.omc-config.json.
Context: This is the primary function of the skill and follows standard practices for CLI tool configuration.
[PROMPT_INJECTION]: Implements a template interpolation system that serves as an indirect prompt injection surface (Category 8).
Ingestion points: Processes data from the local configuration file and session metadata (e.g., sessionId, reason, question).
Boundary markers: No explicit delimiters or instructions are provided to the agent or the notification service to distinguish between template structure and interpolated untrusted data.
Capability inventory: Triggers network requests (curl) or local CLI commands based on the configured events.
Sanitization: While the skill mentions template validation, it lacks specific escaping or filtering for content being placed into templates (e.g., {{question}} or {{reason}}).

configure-notifications