Conciseness

The body is largely efficient — guardrails, signal gathering, and tracing steps all earn their tokens — but there is trimmable duplication: the four reference files are listed twice (Signal Routing table plus the Reference Files section), and the sibling-function instruction appears in When to Use, Step 3, Step 4 (with three examples), and the Checklist. This fits anchor 4 (efficient with minor instances that could be trimmed) better than anchor 5, where every token earns its place.

Actionability

Concrete, copy-paste-ready commands are present ("npx convex insights --details", the "npx -y convex@latest" fallback, and deployment flags), and the tracing steps name exact API calls ("ctx.db.get()", "ctx.db.query()", "ctx.db.patch()", "useQuery", "usePagitatedQuery"-style client hooks). Specific numbered verification criteria complete the executable guidance; as an instruction-only skill with guidance this concrete, it matches the top anchor.

Workflow Clarity

A five-step numbered workflow with an explicit scoping step, a signal-gathering decision tree, an escalation checkpoint before invasive fixes, and a dedicated "Verify before finishing" step listing five concrete confirmations plus an end checklist. This matches the score-5 anchor (clear sequence, explicit validation steps, checklist for complex processes).

Progressive Disclosure

The body is an overview that routes symptoms to four one-level-deep reference files ("references/hot-path-rules.md", "references/occ-conflicts.md", "references/subscription-cost.md", "references/function-budget.md"), all of which exist in the bundle and cross-reference only each other. The routing table plus annotated reference list makes navigation easy, matching the score-5 anchor; the duplicated reference listing is a conciseness issue, not a structural one.

Specificity

The description names the domain ("Audits Convex performance") and enumerates four concrete audit areas — "reads, subscriptions, write contention, and function limits" — which comprehensively cover the skill's scope as reflected in its four reference files. It matches the anchor for multiple specific concrete actions with comprehensive coverage rather than the score-4 anchor, which requires minor gaps.

Completeness

It explicitly answers both what ("Audits Convex performance for reads, subscriptions, write contention, and function limits") and when ("Use for slow features, insights findings, OCC conflicts, or read amplification") with concrete trigger phrases. This is a direct match for the score-5 anchor exemplar; nothing is merely implied.

Trigger Term Quality

"Use for slow features, insights findings, OCC conflicts, or read amplification" includes natural phrases like "slow features" and product-specific "insights findings", but leans technical — missing common variations users would say like "Convex is slow", "expensive queries", or "bytes read". Good keyword coverage with a few natural terms missing fits the score-4 anchor; it is clearly above anchor 3 (missing common variations) but short of anchor 5 (comprehensive synonyms).

Distinctiveness Conflict Risk

"Convex performance" auditing is a clear niche with distinct, product-specific triggers (OCC conflicts, insights findings, read amplification) that would not fire for generic debugging or setup skills. Minimal conflict risk matches the score-5 anchor.