Conciseness

The body is dense, imperative bullets with zero padding — no explanations of concepts Claude already knows, every line a project-specific invariant, command, or file path (e.g. "Upsert events by local `eventIndex`; on final snapshot, delete rows beyond the final event length"). It clearly matches the level-5 anchor 'Lean and efficient; assumes Claude's competence; every token earns its place'; it is not level 4 because there is no over-explanation to trim.

Actionability

Guidance is highly concrete: a copy-paste ripgrep command, exact route paths ("POST /eve/v1/session/:sessionId"), exact file conventions ("agent/tools/<snake_case>.ts"), and verification commands ("pnpm typecheck", "pnpm build"). Per the code_vs_instruction note the absence of code is not penalized, but a few directives remain judgment-level rather than executable — e.g. "Consider a turn settled on an event `endsTurn` from `eve/client` accepts" names the events without a minimal code pattern for the check — which keeps it below the fully copy-paste-ready level-5 anchor.

Workflow Clarity

The sections impose a clear order — First Pass (read docs/search before code), then Core Model, Persistence, Streaming, Connections, and a Verification section with explicit checkpoints ("Run `pnpm typecheck` ... `pnpm build` before opening a PR", plus a manual-flow checklist). It falls short of the level-5 anchor because there are no explicit feedback loops (validate → fix → re-run) and the manual verification list lacks pass/fail criteria, matching 'Clear sequence with most checkpoints present; minor validation gaps'.

Progressive Disclosure

With no bundle files present, the ~55-line body is appropriately self-contained: well-organized sections, each a compact set of invariants, with clearly signaled one-level-deep pointers to the project's deeper docs ("Read `AGENTS.md` first", "Read `docs/how-the-chatbot-works.md`", "docs/setup-and-deploy.md"). This matches the level-5 anchor's clear overview with well-signaled references and easy navigation; splitting this little content into separate files would add indirection, not value.

Specificity

The only stated action is the generic "Work safely on Eve-specific internals" — a single vague verb — while the rest of the description enumerates areas/objects ("withEve routes", "persisted event logs", "Vercel Connect MCP connections") rather than concrete actions the skill performs. It matches the anchor 'Names domain and 1-2 concrete actions, but not comprehensive' more than the level-4 anchor, which requires several specific actions; it is above level 2 because the domain and touched surfaces are named precisely.

Completeness

Both parts are present: 'what' ("Work safely on Eve-specific internals in the Eve chat template") and an explicit, rich 'when' clause ("Use when changing the Eve agent, withEve routes, ...") with concrete trigger phrases. Not a 5 because the 'what' is generic — "work safely" never states what the skill actually does (preserve protocol/durable-session invariants) — placing it at the level-4 anchor where one half is weaker than the other.

Trigger Term Quality

The 'Use when' clause lists many identifiers a developer would naturally mention ("useEveAgent", "ClientSession/SessionState", "withEve routes", "stream/resume behavior", "Vercel Connect MCP connections"), giving good keyword coverage. Not a 5 because common synonyms and variations are missing — e.g. "durable session", "eveSession", "chat_event", "turn.settled", or the stream endpoint path — which the body shows are central vocabulary for this skill.

Distinctiveness Conflict Risk

The description is anchored to a clear niche — Eve-specific internals with named identifiers ("eveChannel auth", "withEve", "useEveAgent") — that no other skill would plausibly claim, giving minimal conflict risk. It clearly matches the level-5 anchor 'Clear niche with distinct triggers'.