Conciseness

The body is largely lean — command blocks, tables, and terse rules of thumb ("**`.commit()`** is required on workflows", "Never bare model IDs") with no explanation of concepts Claude already knows. It falls short of the score-5 anchor because of avoidable padding: the package-to-purpose mapping appears twice (the table under "Bundled docs in `node_modules/`" and again under "Mastra Package Map", both deferring to references/packages.md), and time-sensitive version pins ("v1.6.2 / core 1.27.0 as of 2026-04", "v2.0.0") sit inline rather than in a deprecated/versions section, which the guidelines penalize. It is clearly above score 3, whose anchor requires genuinely unnecessary explanation.

Actionability

Guidance is copy-paste ready throughout: install commands ("pnpm add mastra", "pnpm add @mastra/core @mastra/memory @mastra/rag"), a canonical lookup flow ("cat node_modules/@mastra/core/dist/docs/SKILL.md", "grep -rli \"Agent\" ...", "cat ... SOURCE_MAP.json | jq '.\"Agent\"'"), typecheck commands ("pnpm tsc --noEmit"), and error-string search commands ("rg -n \"error string fragment\" node_modules/@mastra/core/dist"). Concrete rules ("Always `provider/model`", "`threadId` + `resourceId` must be stable") plus executable recipes in references/agents-and-workflows.md cover the common cases, matching the score-5 anchor; there is no pseudocode or vagueness that would pull it to 4.

Workflow Clarity

A clear sequence with explicit validation and a feedback loop: Prerequisites (verify install → install only what's needed) → the numbered six-step discipline in "Critical: Do Not Trust Internal Knowledge" (resolve version → read shipped SKILL.md → verify symbols → "Run typecheck after every change" → enumerate, never invent → surface deprecations) → "When Typecheck or Runtime Fails" provides the error-recovery loop (grep common-errors.md for the symptom, then rg installed source, fix, re-typecheck). This matches the score-5 anchor (explicit validation steps, feedback loops, checklist-style numbered process); the operations are not destructive/batch, so no cap applies.

Progressive Disclosure

The body is a genuine overview that defers detail through three well-signaled, one-level-deep references — each linked with a description of exactly what it contains ([`references/packages.md`](references/packages.md) "every Mastra package... the exact `dist/docs/` path", [`references/agents-and-workflows.md`](references/agents-and-workflows.md) "canonical recipes", [`references/common-errors.md`](references/common-errors.md) "symptom → cause → fix") — and all three files exist in the bundle with the promised content. Navigation is easy and the split is appropriate (lookup paths / recipes / error index), matching the score-5 anchor; score 4 would require organization gaps, which are absent.

Specificity

The description lists multiple concrete capabilities — "Answer questions about the Mastra AI framework and help build agents, workflows, tools, memory, and RAG features", "run the `mastra` CLI", "integrate `@mastra/*` packages", "debug TypeScript errors from Mastra APIs or model-router strings" — comprehensively covering the framework's surface. This matches the score-5 anchor (multiple specific concrete actions, comprehensive coverage); it is not score 4 because there are no meaningful coverage gaps, and not below because nothing is generic.

Completeness

Both questions are answered explicitly: the "what" ("Answer questions about the Mastra AI framework and help build agents, workflows, tools, memory, and RAG features") and a four-part numbered "Use when developers: (1)... (2)... (3)... (4)..." clause plus a concrete "Triggers on:" phrase list. This is the score-5 anchor verbatim in structure (clear what AND when with concrete trigger phrases), well above the score-4 anchor where the "when" is only loosely explicit.

Trigger Term Quality

The explicit trigger list — "Mastra", "@mastra/core", "mastra agent", "mastra workflow", "createWorkflow", "createTool", "createStep", "mastra memory", "semantic recall", "mastra rag", "mastra deployer", "mastra evals", "mastra studio", "provider/model" — covers the natural phrasings a user would say for this domain, including both feature-level names ("semantic recall") and API-level names ("createWorkflow"). It fits the score-5 anchor (comprehensive natural terms including synonyms); score 4 would require noticeable missing natural variations, which are absent.

Distinctiveness Conflict Risk

The description occupies a clear niche (a specific named framework with package-scoped triggers like "@mastra/core" and "mastra studio"), giving minimal conflict with unrelated skills. However, the SKILL.md body itself acknowledges a sibling `mastra` skill covering the same framework with overlapping trigger words ("Mastra", "mastra agent"), so trigger-level overlap with that closely related skill is a real, if minor, risk — the score-4 anchor ("mostly distinct; minor overlap risk with closely related skills") is the best fit, not score 5's "minimal conflict risk".