Conciseness

The body is lean throughout: no explanations of concepts Claude already knows, no padding, and every section (Inputs, Output, Convergence, Anti-patterns, Status) carries unique operational information. It matches anchor 5 — every token earns its place; even the rationale paragraph is scoped to a spec citation rather than generic background.

Actionability

Guidance is concrete: named input files ('plan/steps.json', 'code/index.json', 'iterations'), an explicit receipt directory layout with the receipt's fields ('{ files: [...], diffSummary, rationale, completedAt }'), a complete jsonc loop-config block, and four specific forbidden behaviours. It is not 5 because details like the receipt's exact schema, the diffSummary format, and how 'smallest possible localised edit' is judged are left unspecified — minor gaps for an instruction-only skill.

Workflow Clarity

The sequence is inferable (read inputs → apply one step → write receipt → update status → repeat under an explicit convergence condition with a hard iteration cap), and the anti-patterns act as guardrails. However, this is a batch/destructive file-editing workflow with no explicit validation or verification checkpoint (e.g., confirm the diff applies cleanly or re-check step ownership on retry), so per the judging guidelines workflow clarity is capped at 3: sequence present, checkpoints missing or implicit.

Progressive Disclosure

The skill is under 50 lines, self-contained, needs no external reference files (no references/, scripts/, or assets/ exist, and none are required), and is organised into clean, well-labeled sections. Per the rubric's simple-skill guidance this matches the top anchor: a clear overview that is easy to navigate with no inline content that belongs in a separate file.

Specificity

The description names its domain ('rewrite-plan step') and two concrete actions — 'Apply one rewrite-plan step at a time as small reviewable file edits' and 'never rewriting whole files when a localised change suffices' — but stops there. It matches anchor 3 (domain plus 1-2 concrete actions, not comprehensive); it is below anchor 4 because it lists no further actions (receipts, status updates, iteration handling) and above anchor 2 because the actions it does name are concrete rather than generic.

Completeness

The 'what' is clear (apply plan steps as small, localised, reviewable edits; never rewrite whole files), but 'when' is entirely absent — there is no 'Use when...' clause or equivalent trigger guidance, which caps completeness at 3 per the judging guidelines. It is not 2 because the 'what' is concrete, and not 4 because the 'when' is not even weakly implied.

Trigger Term Quality

The only candidate keywords are internal jargon ('rewrite-plan step', 'localised change', 'reviewable file edits') — terms a user would not naturally say when needing this skill, and there are no synonyms or plain-language variations. This fits anchor 2 (keywords present but missing the natural phrases users say); not 1 because the terms are at least domain-relevant rather than entirely generic, and not 3 because no common user-facing variation is covered.

Distinctiveness Conflict Risk

'Apply one rewrite-plan step at a time' carves a clear niche tied to a specific plan-driven workflow, making confusion with general editing skills unlikely. It is not 5 because the description has no explicit triggers, so selection depends entirely on the 'what' statement; it is clearly above 3 because 'rewrite-plan step' is far more specific than the broad overlap examples at that level.