The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill creates an observability pipeline that ingests and processes untrusted agent conversation data and tool outputs for tracing purposes.
Ingestion points: Conversation history and tool arguments are captured within the agent_tracing.py and agent-tracing.ts files (specifically through the history and arguments parameters).
Boundary markers: There are no explicit prompt isolation or data sanitization mechanisms implemented in the tracing logic to distinguish instructions from diagnostic data.
Capability inventory: The skill enables tool execution wrappers and exports collected telemetry data to the external endpoint https://ingest.maple.dev.
Sanitization: The skill provides user configuration via the OTEL_INSTRUMENTATION_GENAI_CAPTURE_MESSAGE_CONTENT environment variable to toggle message content capture.
[EXTERNAL_DOWNLOADS]: The skill requires the installation of standard OpenTelemetry and SDK instrumentation packages from official registries.
Evidence: The documentation instructs the user to install packages such as opentelemetry-sdk, opentelemetry-exporter-otlp-proto-http, and SDK-specific instrumentors like opentelemetry-instrumentation-genai-openai.
[COMMAND_EXECUTION]: The skill provides logic to execute tool functions using model-generated inputs.
Evidence: The run_tool function in references/python.md dynamically calls agent tools by unpacking arguments parsed from model output using tool(**json.loads(arguments)).

maple-agent-tracing-provider-sdks