The Agent Skills Directory

CREDENTIALS_UNSAFE (LOW): Hardcoded database credentials (password="password" and adminpassword) are present in examples/metrics-dashboard-backend/api.py and the setup documentation in README.md. While common for local development examples, these represent a security risk if used in production environments.
PROMPT_INJECTION (MEDIUM): The skill establishes an ingestion pipeline for untrusted external data that lacks sanitization, creating a surface for Indirect Prompt Injection.
Ingestion points: The MQTT subscriber in examples/iot-data-pipeline/main.go reads raw JSON payloads from the sensors/# topic.
Boundary markers: No delimiters or "ignore embedded instructions" warnings are used when processing or storing the sensor data.
Capability inventory: An agent utilizing this skill to monitor metrics might have downstream capabilities (e.g., triggering alerts, modifying system state, or summarizing logs). Attackers publishing to the MQTT broker could inject instructions into fields like location or sensor_id to influence the agent's behavior.
Sanitization: The Go pipeline and FastAPI backend perform standard JSON unmarshaling but do not sanitize string values against potential injection payloads.
DATA_EXPOSURE (LOW): The FastAPI application in examples/metrics-dashboard-backend/api.py is configured with overly permissive CORS settings (allow_origins=["*"]), which could allow unauthorized cross-origin requests in a web environment.

using-timeseries-databases