Conciseness

The body is dense and assumes competence — no concept explanations (nothing teaches what AVFoundation or a haptic is), and the Anti-Rationalization table carries genuinely non-obvious, hard-won knowledge ("its *format* differs per device for the same library (`i.…` on one, bare numeric on another)", "a suspension never ended hangs the whole group"). It stops short of 5 because the Quick Reference table, the numbered Decision Tree, and Example Invocations restate largely the same ~25 routing mappings three times — a real trimming opportunity even if intentional redundancy for a router.

Actionability

Directives are concrete and specific: "Read: `skills/camera-capture.md` (Pattern 8, deferred start)", "Launch `camera-auditor` agent", and API-level instructions like "hand-rolling reinvents startup barriers... `AVDelegatingPlaybackCoordinatorPlayCommand.hostClockTime` gives you an absolute `CMClockGetHostTimeClock()` start time". It sits at 4 rather than 5 because resolving the `skills/*.md` paths depends on suite context (the inline block explains this for one agent only), and a few entries point at pattern/section numbers without a fallback when the file isn't found.

Workflow Clarity

The routing workflow — identify the media feature, then read the matching reference — is unambiguous and reinforced by three complementary structures (symptom table, decision tree with a dot graph, numbered list). The Cross-Domain Routing section explicitly disambiguates every multi-domain case with "stay here" vs "invoke axiom-X", and Example Invocations map common user phrasings to actions. No destructive or batch operations exist, so the validation cap doesn't apply; the single action is unambiguous, matching the simple-skill exception.

Progressive Disclosure

The body is a pure, well-organized index: ~25 one-level-deep `skills/*.md` references, each clearly signaled with what it covers ("Camera freezes, black preview, rotation → See `skills/camera-capture-diag.md`"). No content that belongs in a separate file is inlined — even the Anti-Rationalization table is router bait (trigger phrases for wrong assumptions) rather than inlined detail. No bundle files were provided to verify against, so this is scored on the body's structure, which matches the 5 anchor: clear overview, well-signaled one-level-deep references, easy navigation.

Specificity

The description lists many concrete work areas — "camera, photos, audio, haptics, ShazamKit, on-device music analysis, face grouping and video highlights, screen recording, casting to non-AirPlay routes, SharePlay playback... CarPlay app design, templates, and navigation" — plus a second sentence of concrete frameworks ("Covers AVCaptureSession, AVFoundation, Core Haptics, MediaPlayer, MusicKit, NowPlaying, ScreenCaptureKit, DockKit"). It falls just below the 5 anchor because it names domains/topics rather than actions, and a few covered areas (video export/playback, DockKit stands) appear only as framework names or not at all in the trigger list.

Completeness

Both halves are explicit: the "what" is stated via "Covers AVCaptureSession, AVFoundation, Core Haptics, MediaPlayer, MusicKit, NowPlaying, ScreenCaptureKit, DockKit" over a concrete domain list, and the "when" opens with "Use when working with camera, photos, audio, haptics, ShazamKit..." — explicit trigger phrases throughout. It clearly matches the 5 anchor rather than 4, where "when" would be less explicit or specific.

Trigger Term Quality

Natural user phrasings are well covered — "camera", "photos", "audio", "haptics", "screen recording", "casting", "SharePlay", "the user's Apple Music library", "lock-screen metadata", "CarPlay" — alongside technical triggers. A few natural terms are missing: "Now Playing" appears only as the framework name "NowPlaying", and "video export/playback", "Chromecast/DLNA", and "gimbal/stand" (DockKit's user-facing vocabulary) are absent, keeping it just below comprehensive.

Distinctiveness Conflict Risk

The Apple-media niche is clear and mostly distinct (ScreenCaptureKit, CarPlay, SharePlay, Core Haptics are unambiguous triggers), but broad terms like "audio" and "music" create minor overlap risk with closely related skills (e.g., speech/transcription or general audio-AI skills, which the body itself routes elsewhere). Not the 5 anchor, which requires minimal conflict risk, but comfortably above 3.