create-honeycomb-board
Pass
Audited by Gen Agent Trust Hub on Oct 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests data from local source files and Honeycomb API results to design dashboards, creating a potential vector for malicious instructions embedded in those data sources.\n
- Ingestion points: Uses file system tools (Read, Grep, Glob) to gather context from code and documentation, and API tools (get_slos, find_queries, run_query) to fetch telemetry metadata.\n
- Boundary markers: No specific prompt delimiters or instructions are provided to isolate and ignore potentially malicious content within the ingested files.\n
- Capability inventory: The agent can create external resources via
mcp__honeycomb__create_boardand execute queries viamcp__honeycomb__run_query.\n - Sanitization: Ingested text is not explicitly sanitized before being placed into board descriptions or text panels.\n
- Mitigation: The impact is mitigated by a non-negotiable human-in-the-loop confirmation step that displays the full proposed board plan to the user for review before execution.
Audit Metadata