The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes a local shell script (send_webhook.sh) to perform configuration checks and transmit data. This script-based approach is used to prevent sensitive access tokens from appearing in system process lists or shell history.
[EXTERNAL_DOWNLOADS]: The skill refers users to https://www.andreas.ink/agent to obtain the necessary publish-only tokens. This domain is an official resource belonging to the vendor (AndreasInk).
[DATA_EXFILTRATION]: The skill manages a sensitive configuration file located at ~/.config/agent-alerts/token.env. It specifically instructs the agent to set restrictive file permissions (mode 0600) and contains strict guardrails to prevent the agent from leaking the token in conversations, logs, or command arguments.
[PROMPT_INJECTION]: The skill defines a pipeline for transforming data from external triggers (monitors, CI jobs, etc.) into notification payloads, creating an indirect prompt injection surface.
Ingestion points: External data enters the context through instructions in SKILL.md for constructing the compact-webhook-payload.json file.
Boundary markers: The skill enforces structural boundaries through JSON schema validation and specific character limits for fields like title and summary.
Capability inventory: The skill utilizes local shell script execution (send_webhook.sh) and outbound HTTPS communication to a specific vendor endpoint.
Sanitization: Mandatory validation steps using --check and --check-config flags are required to verify payload integrity before any transmission.

agent-alerts-setup