generate-titlelisted

# generate-title One title per clip. Third-person, names the subject, promises a specific moment, ALL CAPS, ≤7 words. Designed for the title-transition skill. ## Invoke ``` .claude/skills/generate-title/generate-title.sh <clip_transcript.json> <ingest.json> <out.txt> [title-context.json] ``` - `clip_transcript`: clip-local transcript with `words[]` and/or `segments[]` - `ingest`: the source video's ingest.json (provides title, uploader, url — helps name the subject when the transcript doesn't say it explicitly) - `out`: text file with the title (single line, ALL CAPS, ≤7 words) - `title-context` (optional): pick-segments' per-span judgment — `{topic, rationale, title_suggestion}` — sliced out of `segments.json` by the orchestrator. The clip-local transcript alone can't carry the speaker's register, so a line read literally can invert the meaning (an ironic "what he refuses to think about" titled as a sincere confession). When present, the model reads tone from this context FIRST (Step 0: sincere / ironic / joking / provocative) and titles the actual point, not the surface words. Never shown to the viewer; omitting it falls back to the old clip-only behavior. ## Output A single-line text file. Example contents: ``` WATCH SPEED RAGE AT STREAM SNIPERS ``` ## Prompt principles (enforced) - Third person — never "I", "me", "my", "we". - Name the subject (inferred from clip transcript + ingest.json). - Promise ONE specific moment, behavior, or reaction — n