Conciseness

The body is lean and telegraphic, assumes Claude's competence, and never pads with known concepts ('Codebases accumulate implicit recurring structures' is the only framing sentence). A few cryptic lines — 'All Rosette prep steps MUST be FULLY completed, load-context skill loaded and fully executed' and 'ACQUIRE reverse-engineering/SKILL.md FROM KB' — are compressed to the point of ambiguity rather than earning their place, keeping it below the anchor-5 'every token earns its place'.

Actionability

Steps are concrete: read CODEMAP.md, 'limit top 10-15 most common patterns', 'limit reading samples to 2-3 files per pattern', per-pattern files with Name/Description/Template, INDEX.md headers formatted as '## Pattern Name - short description' and CHANGES.md as '## [YYYY-MM-DD] Brief changes made'. This is executable guidance with minor gaps — the KB acquisition step and the 'Would we rebuild this?' test's application are underspecified — matching anchor 4 rather than the fully copy-paste-ready anchor 5.

Workflow Clarity

A clear 6-step sequence with mode handling (CHECK-EXISTS, IDENTIFY-GAPS, CREATE-OR-UPDATE, PRESERVE-HUMAN, REPORT-CHANGES) and a closing validation_checklist ('No human-curated content overwritten in upgrade mode') that covers the batch-write risk. The dual-mode terms are introduced mid-step rather than defined upfront, a minor checkpoint gap consistent with anchor 4; the explicit checklist and preserve-guard keep it above anchor 3.

Progressive Disclosure

The body is under 50 lines, organized into clearly delimited tagged sections (role, when_to_use_skill, core_concepts, process, validation_checklist), and needs no bundle files. It drops to anchor 4 rather than 5 because it references external resources — 'ACQUIRE reverse-engineering/SKILL.md FROM KB' and the 'load-context' skill — that are neither part of the bundle nor clearly signaled, and 'if state.composite = true' introduces undeclared state.

Specificity

The description 'Extract code patterns.' names the domain but offers a single generic action, matching the anchor example 'Processes PDF files' at level 2. It does not reach level 3, which requires 1-2 concrete actions (e.g., writing pattern files, building an INDEX, analyzing recurring structures) beyond the one vague verb.

Completeness

It has a vague 'what' (extract what, into what form?) and no 'when' at all — there is no 'Use when...' clause or equivalent trigger guidance, which per the judging guidelines caps completeness at 3. This matches anchor 2 exactly ('Has a vague what and no when'); it does not reach level 3 because the 'what' remains too generic to count as clear.

Trigger Term Quality

'code patterns' is a natural, relevant keyword a user might say, fitting anchor 3 ('some relevant keywords but missing common variations or synonyms'). It falls short of anchor 4 because no synonyms or variations appear — conventions, templates, recurring structures, codebase standards, or reverse-engineering terminology.

Distinctiveness Conflict Risk

'Extract code patterns' is somewhat specific but could plausibly trigger for refactoring, code-review, or codebase-documentation requests, matching anchor 3. It is more specific than the level-2 'Helps with document files' but lacks the distinct trigger phrases needed for level 4.