The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill creates an instrumentation layer that ingests and processes agent inputs, outputs, and tool results, which are considered untrusted data from an agentic perspective.
Ingestion points: The skill captures data at the agent.run() and workflow.run() boundaries using the openinference-instrumentation-llama-index package.
Boundary markers: The instructions do not define specific prompt delimiters for the captured data, although this is typical for telemetry instrumentation.
Capability inventory: The skill possesses the capability to export all captured data to external network endpoints (ingest.maple.dev) via the OTLPSpanExporter.
Sanitization: The skill documents the ability to use hide_inputs and hide_outputs configurations within TraceConfig to prevent sensitive content from being captured.
[EXTERNAL_DOWNLOADS]: The skill requires the installation of external dependencies for its tracing functionality.
Evidence: The documentation specifies installing llama-index-core, openinference-instrumentation-llama-index, opentelemetry-sdk, and opentelemetry-exporter-otlp-proto-http via standard package managers.
Context: These are well-known, industry-standard libraries for AI framework development and OpenTelemetry observability.

maple-agent-tracing-llamaindex