Conciseness

The ~30-line body is dense and lean: every line directs action ("Replace every `TODO(agent)` marker", "Keep slugs stable across runs", "Plans go to `<plansDir>/<run>/<slug>.md`") with zero padding and no explanation of concepts Claude already knows. Clearly matches the anchor-5 example of every token earning its place.

Actionability

Provides a copy-paste-ready scaffold command with enumerated flags and area values (`pnpm exec tsx .agents/skills/refactor-plan/scripts/new-plan.ts --run <id> --slug <kebab> ... --area <Framework|Slides|...>`), the exact output location, concrete bug-plan variant behavior, and an explicit done criterion (findings-upsert.ts acceptance). Delegates detail to real, existing reference files — fully executable guidance matching anchor 5.

Workflow Clarity

Clear sequence across sections — judge first (rubric.md, read diffs, find a named repeated mechanism), scaffold (command), fill (replace markers per plan-guide.md), with an explicit end-state validation ("A plan is finished when fragility-common's `findings-upsert.ts` accepts it"). The recovery loop when findings-upsert.ts rejects a plan (fix markers/confidence, re-run) is only implied via the reference rather than stated in the body, which is the minor checkpoint gap of anchor 4 rather than the explicit feedback loop of anchor 5.

Progressive Disclosure

The body is a concise overview with two well-signaled, one-level-deep references — "`references/rubric.md` separates fragile from fast-moving from coupled, and sets confidence" and "following `references/plan-guide.md`" — both of which exist in the bundle and carry the detail, while executable logic lives in `scripts/new-plan.ts` (with `lib.ts`). No nested doc chains, content appropriately split; matches anchor 5.

Specificity

Names concrete actions — "Judge whether a system is fragile or just fast-moving" and "write a ticket-ready refactor plan for one root problem" — with specific deliverable attributes ("measured evidence", "a stable fingerprint"). Falls between anchor 3 (1-2 concrete actions, not comprehensive) and anchor 5 (comprehensive coverage); it is more specific than 3 but does not enumerate the full range of capabilities, so 4 fits.

Completeness

Explicitly answers both questions: what ("Judge whether a system is fragile or just fast-moving, and write a ticket-ready refactor plan... with measured evidence and a stable fingerprint") and when ("Use after system history (or a bug trace) points at a systemic problem worth fixing at the root") with concrete trigger conditions. The 'when' clause is explicit and specific, so it exceeds anchor 4 ('when' could be more explicit) and matches anchor 5.

Trigger Term Quality

Contains natural domain terms a user would say — "refactor plan", "fragile", "system history", "bug trace", "systemic problem", "root" — but misses common synonyms and variations such as "tech debt", "keeps breaking", or "repeated bugs". Good coverage without being comprehensive, matching anchor 4 rather than 5.

Distinctiveness Conflict Risk

Carves a clear niche — fragility triage that produces a deduplicated, ticket-ready root-cause plan — with a distinct trigger gated on prior system history or a bug trace. Minor theoretical overlap with a generic refactoring skill, but the trigger and deliverables (fingerprint, measured evidence) are distinct enough to match anchor 5 over anchor 4.