The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill defines a logging interface in lib/logger.ts that accepts a context object containing arbitrary data. This data is often sourced from external, untrusted inputs such as HTTP request bodies or API responses. If an AI agent later processes these logs, the embedded data could contain instructions designed to influence the agent's behavior.\n
Ingestion points: The context parameter in Logger methods (debug, info, warn, error) and the child logger method in lib/logger.ts.\n
Boundary markers: Production logs are formatted as JSON, providing structural separation between log metadata and contextual data.\n
Capability inventory: The logger outputs to the system stdout via console.info and can optionally send data to Axiom via the @axiomhq/js library.\n
Sanitization: The implementation automatically prefixes keys that collide with reserved log fields (e.g., timestamp becomes ctx_timestamp), which protects the log schema but does not filter the content of the values.\n- [EXTERNAL_DOWNLOADS]: The skill provides instructions for direct integration using the @axiomhq/js package from the official npm registry. Axiom is a well-known observability and logging service.

structured-logging-axiom