Conciseness

The Workflow steps are dense, assume Convex knowledge (`.collect()`, `.withIndex`, O(table) growth), and never explain basics — but the "## Rules" section substantially restates the workflow: rule 1 repeats step 3 ("Cost = data-read-per-call × call-volume — always show both factors"), rule 5 repeats step 6's confirm-cost gate, rule 3 repeats step 7's "cite the pricing page" caveat, and rule 6 repeats step 1's guard scope. Fits anchor 3 (mostly efficient but could be tightened); not 2 because nothing explains concepts Claude already knows; not 4 because the Rules/Workflow duplication is real, trimmable padding.

Actionability

Concrete guidance throughout: named MCP sources ("`insights` for the bytes-read / documents-read events", "`tables` for row counts", "`functionSpec`"), specific Convex patterns ("index + `.withIndex` instead of scan, `.paginate`/`.take` instead of `.collect`"), and concrete output expectations ("your spend is dominated by messages:list reading the whole table every call; index it and it drops ~100x"). Fits anchor 4 (mostly executable guidance with minor gaps); not 5 because the "cost-class finding on the bus" format is referenced but never shown and the official MCP invocation is not given; not 3 because the named tools and fix patterns make the steps directly executable for an instruction-only skill.

Workflow Clarity

Seven explicitly labeled, well-sequenced steps (GUARD → GATHER → ATTRIBUTE → PROJECT → NAME FIX → CONFIRM-COST → REPORT) with a precondition check (deploy-guard), an explicit gate ("STATE the price and recurrence explicitly and get an explicit yes BEFORE proceeding"), and an error-recovery fallback ("If there's no usage/traffic yet, say so and estimate from the query SHAPES instead"). The workflow is read-only ("Read-only over dev/prod"), so the destructive/batch validation cap does not apply; every checkpoint that matters for a read-only analysis flow is explicit. Not 4 because there is no material validation gap — guard, confirm-cost gate, and the no-traffic fallback are all present.

Progressive Disclosure

The body is under 50 lines, no bundle files exist (references/, scripts/, and assets/ directories are all absent), and none are needed — no external file paths are referenced that could dangle. Content is organized into two clean sections (## Workflow, ## Rules). Per the rubric's simple-skill guideline, an under-50-line skill with no need for external references scores 5 with just well-organized sections, which this has.

Specificity

Quotes: "rank functions by bytes/documents-read × call-volume from insights", "project each cost driver's growth curve", "name the cheapest fix", "confirm-cost for paid actions" — four distinct concrete actions covering the full capability. Matches anchor 5 (multiple specific concrete actions, comprehensive coverage); not 4 because the action list spans the whole workflow (gather→rank→project→fix→confirm) with no gaps rather than only 'minor gaps in coverage'.

Completeness

The 'what' is explicit and multi-part ("rank functions… project each cost driver's growth curve… name the cheapest fix; confirm-cost for paid actions"), but there is no "Use when…" clause or equivalent trigger guidance — the 'when' is only weakly implied by "Preview Convex spend". Per the judging guidelines a missing when-clause caps completeness at 3; not 4 because the 'when' is absent entirely rather than merely imprecise.

Trigger Term Quality

Natural terms present: "Convex spend", "cost driver", "cheapest fix", "paid actions" — phrases a user worried about cost would actually say. Fits anchor 4 (good keyword coverage, a few natural terms missing); not 5 because common synonyms like "billing", "budget", "how much", or "price" are absent and the phrasing leans technical ("bytes/documents-read × call-volume", "insights"); not 3 because the core natural terms (spend, cost, cheapest fix) are present and on-target.

Distinctiveness Conflict Risk

Quotes: "Preview Convex spend", "from insights", "bytes/documents-read" — the niche is explicitly the Convex platform's cost analysis, a clear slot with distinct triggers and minimal conflict risk (it would only fire for Convex cost/spend questions). Not 4 because no meaningful overlap with adjacent perf- or billing-skills is implied by the description itself.