The Agent Skills Directory

[PROMPT_INJECTION]: Indirect prompt injection surface identified in logging reference guides.
Ingestion points: references/flask-integration.md logs the X-Request-ID header and order_id URL parameter; references/express-http-request-logging.md logs x-request-id and userId from request headers and parameters.
Boundary markers: Absent; untrusted external data is interpolated directly into log fields without delimiters or "ignore" instructions.
Capability inventory: The skill's primary function is to create logs. If these logs are later consumed or analyzed by an AI agent, they provide a vector for indirect instruction injection.
Sanitization: Absent; the examples show direct use of request-derived strings without validation or escaping.
[COMMAND_EXECUTION]: Security anti-pattern in reference code logic.
references/flask-integration.md includes a code example for a database query that is vulnerable to SQL injection: db.query(f'SELECT * FROM orders WHERE id = {order_id}') uses an unparameterized f-string with a URL-derived variable.
[EXTERNAL_DOWNLOADS]: Fetches configuration and images from a well-known service.
references/elk-stack-setup.md contains a Docker Compose configuration that pulls official Elasticsearch, Logstash, and Kibana images from docker.elastic.co (Elastic's official registry).

application-logging