Lean, opinionated guidance that assumes Claude's competence — no padding explaining what traces/spans or queries are; every section adds Honeycomb-specific judgment Claude would not already have. Dense but not verbose.

Provides concrete, copy-paste-ready expressions and tool calls throughout — e.g. `COUNT, P99(duration_ms), HEATMAP(duration_ms)`, `MUL(IF($error, 1, 0), 100)`, `any.service.name`, and named MCP calls like `find_queries`, `find_columns`, `run_bubbleup`.

The "Before Every Query" section is an explicit pre-query checklist with a validation checkpoint ("Validate columns with find_columns before querying"), plus a decision table and ordered principles; reads are non-destructive so no retry loop is required.

The body is a concise overview that points to five one-level-deep reference files (visualize-operations.md, relational-fields.md, query-examples.md, result-interpretation.md, calculated-fields.md), all of which exist in ./references/ and are each clearly signaled with a one-line description.

Lists multiple concrete capabilities with specifics — "operation selection (percentiles not AVG, HEATMAP for distributions), relational field patterns (root., parent., any., none.), calculated fields, query math, and result interpretation (P99/P50 ratios, heatmap bands, TOTAL/OTHER rows)" — far beyond naming a domain.

Explicitly answers both what (constructing/interpreting Honeycomb queries across named sub-areas) and when via a clear "Use this skill when the user wants to query spans, traces, or log/event data" clause with concrete triggers.

Embeds natural phrasings users would actually say — "show me latency", "error rate", "find slow requests", "find outliers", "interpret results", "download raw results" — giving good coverage of natural trigger terms.

Niche is sharply scoped to Honeycomb trace/event queries and explicitly excludes metrics datasets ("for those, use metrics-queries instead"), actively disambiguating from a sibling skill.