Conciseness

The body is dense with skill-specific operational knowledge Claude does not have (exact flags, safe-zone percentages, font-coverage table, backend selection rules) and explains no general concepts. Minor trims are possible — the look descriptions partially repeat the frontmatter, and the time-sensitive "verified by rendering, 2026-07-27" date sits outside any old-patterns section — which keeps it at efficient-with-minor-trims rather than anchor 5.

Actionability

Every step is a copy-paste-ready command with full paths, exact flags (--max-words, --bridge, --tail, --font-key, --maxw-frac), fallback rules (OpenAI STT vs faster-whisper by env key), and an explicit decision tree for routing. This matches the anchor of fully executable commands covering the common cases, with the worked example for each of the three looks.

Workflow Clarity

The sequence is clearly ordered (route selection → Whisper timings → authored-wording swap → burn → replace deliverable) with a preflight import check and an explicit failure fallback (deliver unsubbed, say so). However, step 4 destructively overwrites the original video file (`mv final_subbed.mp4 <original name>`) with no verification of the burned output beforehand, so per the rubric's destructive-operation cap this cannot score above 3 despite the otherwise strong sequence.

Progressive Disclosure

Structure is good: the body is a lean overview of routes, pipeline, hard rules and defaults, while the three referenced scripts (audio_to_captions.py, subtitle_paper_burn.py, burn_caps_clean.sh) exist in the bundle and carry the implementation detail. Minor gaps: the body also references a fonts/ subdirectory that is not present in the bundle's scripts/, and some content (the font-coverage table, safety notes) could live in a reference file.

Specificity

Concrete actions are named — "Burn timed subtitles onto a finished video or configure Whisper-timed caption burning during faceless-video assembly" and "returns a captioned video or the exact backend subtitle configuration" — plus the three parameterized looks (paper/bold/clean) with visual specifics. It stops short of the anchor-5 example's breadth of distinct actions (essentially two verbs here, burn and configure), so anchor 4 with minor coverage gaps fits best.

Completeness

Both halves are explicit: the what is "Burn timed subtitles onto a finished video... returns a captioned video or the exact backend subtitle configuration", and the when is a dedicated "Use when: adding/burning subtitles or captions to an existing video, restyling captions, or when a production workflow needs its finished cut subtitled" with concrete trigger phrases — matching the anchor-5 example's structure exactly. It even adds an explicit NOT-for clause, which exceeds the anchor.

Trigger Term Quality

Natural phrases users would say are present: "adding/burning subtitles or captions to an existing video", "restyling captions", "needs its finished cut subtitled". Common synonyms like "closed captions", "CC", "hardcode", or the .srt extension are absent, which keeps it at good-but-not-comprehensive coverage rather than anchor 5.

Distinctiveness Conflict Risk

A clear niche (burning Whisper-timed captions in three named looks) with distinct triggers and an explicit "NOT for: generating the video, translating speech, or live/soft subtitle tracks" boundary. This is far more fenced-off than the anchor-4 "minor overlap risk" case and cleanly establishes a minimal-conflict trigger set.