Conciseness

The body is dense and assumes competence — it explains only OpenMontage-specific and Azure-specific facts (fallback chain, cost conventions, determinism), never generic TTS concepts. Not 5 — the opening paragraph restates the frontmatter description nearly verbatim ("Optional cloud TTS provider — ... piper_tts remains the default offline path") and could be trimmed.

Actionability

Copy-paste-ready registry call with real parameter values (voice alias, "-4%" rate, output_path/format), executable env-var setup, a concrete curated voice table, concrete param ranges, and per-call cost math. Not 4 — the common case is fully covered; locale and style are advanced options and are flagged as such.

Workflow Clarity

A single-action skill whose flow is unambiguous and presented in execution order (setup env → availability gate → generate per segment → fallback chain), and batch operations carry explicit validation checkpoints ("listen to the first generated segment before batch-running a full script", "listen to a sample before batch runs"), so the batch cap does not apply. Not 4 — the verification steps are explicit, not merely implied.

Progressive Disclosure

No bundle files exist and none are needed: the curated five-voice shortlist, parameters, and cost notes are appropriately sized for SKILL.md, sections are clearly organized, and the only external links (MS REST docs, voice gallery, pricing) are one level deep and well-signaled. Not 4 — nothing inlined here would be better split into a separate file at this size.

Specificity

"Generate neural narration audio using Azure AI Speech (REST text-to-speech)" names a precise action and mechanism, and "Shares one Speech resource with azure_stt" adds a concrete setup capability. Not 3 — the action is specific rather than generic like "Processes PDF files"; not 5 — only one core action is listed rather than several.

Completeness

Both what ("Generate neural narration audio using Azure AI Speech (REST text-to-speech)") and when ("Use when synthesizing voiceovers or narration in OpenMontage") are explicit, with concrete trigger phrases and an explicit preference condition (AZURE_SPEECH_KEY configured). Not 4 — the when clause is concrete rather than vague.

Trigger Term Quality

Natural trigger terms include "voiceovers", "narration", "text-to-speech", and "TTS", giving good synonym coverage. Not 5 — common phrasings like "voice", "speech", or "audio" as standalone triggers are missing.

Distinctiveness Conflict Risk

It explicitly disambiguates from siblings ("the local piper_tts remains the default offline path", "Shares one Speech resource with azure_stt"), but the "voiceovers or narration" triggers equally match other TTS skills (elevenlabs_tts, openai_tts), leaving minor overlap risk. Not 5 for that reason; not 3 — the differentiation is explicit, not incidental.