The Agent Skills Directory

[SAFE]: The skill implements standard observability workflows for Honeycomb. It uses authorized tool sequences (context priming, BubbleUp differentiation, and trace inspection) to reach root causes for production issues like latency spikes and error surges.
[INDIRECT_PROMPT_INJECTION]: The skill defines a surface for processing untrusted data (telemetry data such as span attributes, exception messages, and log bodies) from external systems.
Ingestion points: Data enters the context through outputs from run_query, get_trace, get_slos, and run_bubbleup tools (found in SKILL.md and investigation-playbooks.md).
Boundary markers: The instructions do not define explicit delimiters or warnings for the agent to ignore instructions embedded in telemetry data.
Capability inventory: The skill uses tools for read-only query execution (run_query, get_trace) and documentation (create_board).
Sanitization: There is no explicit sanitization of field values (e.g., exception messages) before they are displayed or used in subsequent analysis steps. This surface is characteristic of the skill's primary purpose and poses no inherent malicious risk.

production-investigation