Conciseness

Noticeably verbose for a router: the ~130-line "Example Invocations" section restates the Decision Tree entry-for-entry (e.g. "User: 'My app's memory usage keeps growing' → See skills/memory-debugging.md" duplicates Decision Tree item 2), "Critical Patterns" re-summarizes sections already described in Routing Logic, and inline version markers ("OS27", "iOS 26", "iOS 13–26", "the 27 cycle") appear throughout rather than being isolated in a versioned/deprecated section. This matches anchor 2 — several padded sections — better than anchor 3, since roughly half the body is a second or third restatement of the same routing map.

Actionability

Routing decisions themselves are concrete and specific ("Memory leak scan → axiom-audit-memory (5-phase semantic audit...)", "xcprof compare --fail-on-regression (non-zero exit)"), but execution is incomplete: the destinations "See skills/memory-debugging.md" cannot be resolved — no skills/, references/, or scripts/ files exist in the bundle — and destination naming is inconsistent, with the same target called "axiom-audit-memory", "memory-auditor (Agent)", and "memory-auditor" in different sections. Matches anchor 3: concrete guidance with missing key details; not anchor 4, whose guidance is fully executable as written.

Workflow Clarity

The Decision Tree is a well-sequenced triage with conditionals and even a feedback loop ("memory-debugging FIRST ... then performance-profiling if memory is fixed but stutter remains") and ordering rules, and the Anti-Rationalization table doubles as error-recovery guidance. Below anchor 5 because of real friction: inconsistent agent names across sections, non-monotonic numbering (10/10a/10b/10c, 14/14a), and no checkpoint telling Claude what to do when a referenced sub-skill file is unavailable.

Progressive Disclosure

The router pattern is sound: every detailed topic is deferred one level deep with clearly signaled references ("See skills/hang-diagnostics.md", "See axiom-build (skills/lldb.md)"), and section headers (When to Use, Routing Logic, Decision Tree, Anti-Rationalization) make navigation easy. Not anchor 5 because the referenced paths are not part of this bundle (no skills/ directory exists alongside SKILL.md), and the same routing map is spread across three parallel sections, so content placement is only 'mostly' appropriate.

Specificity

Names the domain ("Apple-platform app") and lists several specific capability areas — "memory leaks, profiling, Instruments workflows, retain cycles" — with only a generic tail ("optimization"). It sits between anchor 3 (1-2 concrete actions) and anchor 5 (comprehensive multi-action list): broader than 3, but the items are topics under 'Covers' rather than explicit actions like 'Diagnose memory leaks' and 'optimization' adds no specificity, so 4 fits better than 5.

Completeness

Both halves are explicit: "Use when an Apple-platform app feels slow, memory grows, battery drains, or ANY performance issue needs diagnosing" (when, with concrete triggers) and "Covers memory leaks, profiling, Instruments workflows, retain cycles, optimization" (what). It falls short of anchor 5 because the 'what' is a bare topic list with the vague 'optimization' rather than the clearly stated concrete actions the anchor-5 exemplar shows; it is clearly stronger than anchor 4's weak/implied 'when'.

Trigger Term Quality

Natural user phrasings are present: "app feels slow", "memory grows", "battery drains", "memory leaks", "retain cycles", "Instruments". Not a 5 because common variations users actually say are missing — 'laggy', 'hangs', 'freezes', 'janky', 'device gets hot' — none of which appear even though the body routes on them.

Distinctiveness Conflict Risk

Clear niche — Apple-platform (iOS/macOS) performance diagnosis with distinctive platform terms ("Instruments workflows", "retain cycles") — and triggers that would not fire for unrelated skills. Matches anchor 5: distinct triggers, minimal conflict risk.