Conciseness

The body is operational throughout — glob patterns, env var names, a deterministic extraction method, and endpoint payloads — and does not explain concepts Claude already knows (no SEO 101 padding). Minor trimmable content keeps it below the top anchor: inline time-sensitive dates ('checked 2026-10-12', twice) sit outside any 'old patterns'/'deprecated' section, and the opening paragraph on ${CLAUDE_SKILL_DIR} resolution is denser than needed. It is clearly above anchor 3, where noticeable unnecessary explanation would be present.

Actionability

Mostly executable guidance: copy-paste-ready endpoint JSON bodies with required parameters, a severity formula with defined variables, skip patterns, credential env vars, and per-error fallback behavior. It is not fully self-contained: API mode 'requires ... a dedicated local CLI wrapper' that is not bundled and has no implementation or invocation instructions (only the SKIPPED fallback), and the local-mode keyword extraction relies on judgment where a precise procedure or script would exist. That places it between anchors 4 and 5, closer to 4.

Workflow Clarity

Both modes have numbered, clearly sequenced steps, and there is a real pre-flight checkpoint — the spend-approval gate ('count the URLs ... Proceed only after the user approves that count, never exceed it') plus retry-once-then-fallback error handling. Minor gaps keep it below 5: Step 4 ('assess severity and generate a recommendation') and Step 5 are underspecified, and API-mode step 1 leaves URL collection vague ('from the user (or sitemap)'). The skill is read-only reporting, so the destructive/batch validation cap does not apply.

Progressive Disclosure

The bundle contains no references/, scripts/, or assets/ directories at all, yet the body's opening directs the reader to 'Read the main reference `orchestration-details.md`' — a dangling reference to a file that does not exist, with the only other paths (${CLAUDE_SKILL_DIR}/../blog/references, etc.) pointing outside the skill. The single-file body also inlines detail (full API endpoint payloads, five recommendation deep-dives) that a well-structured bundle would split into one-level-deep reference files. It has real section structure, so it sits at anchor 3 rather than 2, but the broken/missing reference and monolithic placement of API detail prevent a higher score.

Specificity

Lists multiple specific concrete actions — 'extracting primary keywords from titles and headings, clustering semantically similar targets, and flagging posts competing for the same search intent' — plus mode/cost constraints and output form ('Outputs a severity-scored report with merge or differentiate recommendations'). Third-person voice is used throughout, so no voice penalty applies; coverage is comprehensive with no vague filler.

Completeness

Clearly answers both 'what' (detect, cluster, flag, severity-scored report) and 'when' with an explicit trigger clause: 'Use when the user says "cannibalization", "keyword overlap", "competing pages"'. Concrete trigger phrases and mode/cost caveats make both halves explicit, matching the top anchor exactly.

Trigger Term Quality

Natural trigger phrases users would actually say are present ('cannibalization', 'keyword overlap', 'competing pages'), which is good keyword coverage. It falls short of the top anchor because common variations and synonyms are missing, e.g. 'competing keywords', 'duplicate content', 'keyword competition', or 'multiple pages ranking for the same keyword'.

Distinctiveness Conflict Risk

Clear niche (keyword cannibalization across blog posts) with explicit negative boundaries — 'Not a site audit (blog-audit) or cluster planning (blog-cluster); it changes no files or redirects' — which sharply reduces mis-trigger risk against sibling SEO skills. Distinct, unlikely-to-collide trigger terms match the top anchor.