The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the installation of standard, well-known instrumentation and SDK packages including crewai, openinference-instrumentation-crewai, and opentelemetry-sdk from public package registries.
[COMMAND_EXECUTION]: Provides commands for package managers (pip, uv, poetry) to install the necessary dependencies for tracing.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to trace applications that ingest untrusted user text and conversation history into LLM prompts (e.g., in the handle_message function). While the skill itself is a passive monitoring tool, the underlying CrewAI application has an inherent attack surface for indirect prompt injection via interpolated task descriptions.
Ingestion points: Untrusted data enters the agent context through the text and history arguments in handle_message and build_crew (SKILL.md).
Boundary markers: The instructions recommend including history and user messages directly in the task description without specific prompt delimiters or sanitization warnings.
Capability inventory: The skill facilitates instrumentation of model calls and tool calls, providing visibility into subprocess execution and network operations triggered by the agent.
Sanitization: No explicit sanitization or escaping of external content is specified in the prompt interpolation logic.
[DATA_EXFILTRATION]: The skill configures an OpenTelemetry exporter to send agent trace data, including prompt content and tool outputs, to the vendor's ingestion infrastructure at ingest.maple.dev. This is the documented and intended functionality of the skill for observability purposes.

maple-agent-tracing-crewai