The body is mostly efficient, but the opening blockquote defining Chaos Engineering and the ISTQB/ISO recoverability mapping explain concepts Claude already knows, so it could be tightened; it is not lean enough for the top anchor.

Concrete executable guidance throughout - explicit formulas ('held_rate = count of PASSED runs / total runs'), specific thresholds (20% above/below, 10 percentage points, 3 consecutive PASSED runs), exact output path, and a fully worked example - copy-paste ready decision logic.

A clear 7-step sequence with an upfront 'How to use' overview, explicit validation checkpoints (hard-reject rule for zero reports, INCOMPLETE flagging for missing fields), and an anti-patterns table providing error-recovery guidance.

Overview in SKILL.md with a single one-level-deep reference (references/drill-fields-and-signals.md) that exists on disk and is clearly signaled by inline links in Steps 1 and 4; field-extraction and signal catalog are appropriately split out.

Lists multiple concrete actions: 'per-experiment hypothesis-held / blast-radius / time-to-detect / time-to-recover, degradation trends across runs, action items, and a stakeholder summary', matching the anchor that lists several specific concrete actions.

Clearly answers both what (aggregates drill verdicts into a trend report) and when with an explicit 'Use when a team has completed one or more chaos drills and needs a structured trend report' trigger, matching the top anchor.

Natural terms a user would say are well covered - 'chaos drills', 'trend report', 'resilience', 'improving, degrading, or stable' - rather than technical jargon, satisfying the anchor for good coverage of natural terms.

Occupies a clear niche (post-hoc multi-drill trend aggregation) with distinct triggers; the body even differentiates it from chaos-experiment-author and live drill runs, so it is unlikely to fire for the wrong skill.