Conciseness

The body is lean and assumes competence — short sections, no padding, no explanation of concepts Claude already knows, and actionable heuristics like 'Watch for always-on-convention language: "the agent keeps getting X wrong", "every time"'. It falls short of the every-token-earns-its-place anchor because the body re-states compressed versions of content that also appears in references/choosing-the-shape.md (primitive selection, skill-vs-rule, generic-vs-native hooks), which could be trimmed or deferred.

Actionability

Concrete if/then decision rules ('If the agent should follow a workflow when a relevant task comes up, use a skill'), an explicit plan template ('state: the problem, the artifacts, and the proposed shape, which primitive each part becomes'), a named handoff target ('Pass the confirmed plan to build-composition'), and a stop condition. As an instruction-only skill this is actionable without code, but no worked example of a finished composition plan is provided — matching the mostly-executable-with-minor-gaps anchor rather than the fully-covering-common-cases anchor.

Workflow Clarity

The sequence is unambiguous and in order: choose the primitive, decide the shape, write the plan, confirm with the user ('Confirm with the user, or make the call and tell them your reasoning'), hand off, and stop ('Stop when there is a confirmed composition plan'). Explicit confirmation and stopping checkpoints are present, and this is a planning skill with no destructive or batch operations requiring validation loops.

Progressive Disclosure

The short body serves as an overview and signals its single one-level-deep reference at exactly the right decision point ('See [references/choosing-the-shape.md](references/choosing-the-shape.md)'), and that file exists in the bundle with the deeper rationale (generic vs native hook tiers, verifier guidance, heuristics). Content is appropriately split between the inline decision rules and the deferred depth, making navigation easy.

Specificity

Lists several concrete actions and outcomes — 'Produces a short composition plan - a single skill for simple cases, or a plugin with the right mix of skills, rules, hooks, MCP servers, and commands' and 'confirms it with the user before anything is built'. It falls short of the comprehensive anchor because 'decide the right shape for the context' is abstract phrasing, but it is well above the 1-2-concrete-actions anchor.

Completeness

Both 'what' (produce a short composition plan choosing between a single skill or a plugin with a specific primitive mix) and 'when' (the explicit 'Use once the problem and artifacts are understood' clause, so the missing-trigger cap at 3 does not apply) are stated. The 'when' is explicit but framed as a sequencing condition rather than concrete user-facing triggers, fitting the both-present-but-could-be-more-specific anchor.

Trigger Term Quality

Domain keywords are present ('skill', 'plugin', 'rules', 'hooks', 'MCP servers', 'commands'), but the trigger phrase 'Use once the problem and artifacts are understood' describes an internal pipeline stage rather than language a user would naturally say. Common user variations like 'build a skill', 'create a plugin', or 'package' are missing, matching the some-relevant-keywords-but-missing-variations anchor rather than the good-coverage anchor.

Distinctiveness Conflict Risk

It carves a clear niche — deciding composition shape before building — and its sequencing trigger ('once the problem and artifacts are understood') distinguishes it from build-stage skills. Minor overlap risk remains with closely related skill-authoring skills (e.g., the 'build-composition' step referenced in the body), so it matches the mostly-distinct anchor rather than the minimal-conflict anchor.