Conciseness

The prose is dense and nearly every sentence carries decision-relevant, repo-specific information Claude could not know (e.g., "64,491 assistant records carry an `effort` field and every single one of them reads `high`"). There are minor instances that could be trimmed — the pedantry about the 9% field and closing flourishes like "Worth stating plainly so nobody assumes a reach this does not have" — so it fits anchor 4 rather than 3, which would require more substantial padding.

Actionability

Guidance is concrete and decision-ready: a full rubric table ("`implementer` | opus · xhigh | ..."), a complexity-to-agent table, hard rules ("Ignore a rating carrying `complexity:unconfirmed`", "If the task does not fit any definition... default to `implementer`"), and an executable grep template. It is not 5 only because the one command carries `<package>`/`<workspace>` placeholders and some checks (how to read a confirmed label) are left implicit; per the rubric's code-vs-instruction note, the absence of code is not itself penalized.

Workflow Clarity

The decision flow is legible — read the task, discard unconfirmed labels, match the rubric, verify the semver delta before choosing, output one name plus one sentence — with the output contract stated up front ("One name from `.claude/agents/`, and one sentence saying why") and a pre-decision check ("Check the semver delta before you choose, do not infer it"). It falls short of 5 because the sequence is distributed across prose sections rather than given as an explicit ordered procedure with checkpoints.

Progressive Disclosure

No bundle files exist, and the ~120-line body is organized into well-labeled sections (What this run must produce / The rubric / Four traps / What consumes this) that keep navigation easy. This matches anchor 4's 'good structure, minor organization gaps'; it is not 5 because some evidence-caveat material (the transcript statistics and the unconfirmed-rating discussion) could live in a separate reference file to slim the main body.

Specificity

The description names the domain ("dispatching a subagent and choosing which one") and one to two concrete actions ("mapping a task and its declared complexity to an agent definition, and so to a model and an effort level"), but does not enumerate the broader capability set. It matches the 'names domain and 1-2 concrete actions' anchor rather than 4, which expects several specific actions listed.

Completeness

Both parts are explicitly answered: the 'what' is mapping a task and its complexity to an agent definition, model, and effort level; the 'when' is explicit with concrete trigger phrasing ("Use when dispatching a subagent and choosing which one", "Invoked by orchestrate when it hands off a task"). It is not the 4 anchor because the 'when' is already specific rather than merely present.

Trigger Term Quality

Natural phrases are present — "dispatching a subagent", "choosing which one", "hands off a task", "should this run on a cheaper model" — covering how a user would phrase the need. A few natural synonyms (e.g., 'delegate', 'assign a task') are missing, so it sits between anchor 3 and 5, noticeably above the midpoint.

Distinctiveness Conflict Risk

It occupies a clear niche — agent/model/effort selection at dispatch time — with triggers ('dispatch', 'orchestrate hands off', 'cheaper model') that no neighboring skill would claim. Conflict risk is minimal, matching the anchor 5 example's profile.