feat(skills): add changelog-video skill for repo-native CC + Codex discovery (#2552)

Packages Jake Moran's changelog-video pipeline (v1, validated end-to-end
by Home on the Jun 23-29 range) as a repo-native skill set that Claude
Code (.claude/skills/) and Codex CLI (.agents/skills/) auto-discover the
moment the repo is opened. No install step; run the skill against a
changelog markdown for a given git range and it produces a lint-clean,
seam-gate-green 1080x1080 MP4 (~45-60s, Annie VO, mock-UI visualizations,
caption rail) end-to-end.

Six skills added byte-identical in both mirror dirs:
- changelog-video (pipeline entry point)
- motion-doctrine (carries seam-stamp.mjs + seam-gate.mjs)
- cut-the-curve, captions-overlay, seam-craft, oversized-cursor

Layout:
- .claude/skills/  - Claude Code project-local auto-discover
- .agents/skills/  - Codex CLI project-local auto-discover (verified via
                     Magi's clean-home Codex 0.144.3 repro; NOT .codex/skills/)

Fonts, animated background (12 MB), house BGM (5 MB), lexicon, and
align-captions ship inside the skill dirs. .gitattributes routes only
.claude/skills/**/*.{mp4,mp3} + .agents/skills/**/*.{mp4,mp3} through
LFS — narrowly scoped so unrelated Player, Studio, registry, and
marketplace media stay put. HeyGen CLI auth is the one credential the
skill needs; Node >= 22, ffmpeg, and headless Chrome are documented
alongside in both READMEs.

.gitignore: rewrites .claude/ and .agents/ blocks to keep agent-installed
skill hygiene while re-including the six repo-native skill dirs plus
README.md.

CI:
- Extends changes.skills filter to match .claude/skills/**,
  .agents/skills/**, scripts/lint-skills.ts, and scripts/check-skill-mirror.mjs.
- New 'Skills: project-native lint + mirror' job runs the extended
  lint-skills.ts (schema-driven; required { name, description } + optional
  { license, allowed-tools, metadata }, name pattern check, description
  length check) plus a new check-skill-mirror.mjs byte-integrity script
  (24 mirrored files must match; README.md deliberately per-CLI).
- Wired into 'bun run lint' locally.

Frontmatter validator:
- Rejects unsupported top-level keys (catches category:-style drift).
- Requires name + description.
- Validates name pattern (^[a-z][a-z0-9-]{0,63}$) and description shape
  (non-empty, <=1024 chars).
- Missing frontmatter block itself is a first-class error.

Also strips unsupported top-level 'category:' frontmatter from Jake's
motion-doctrine and cut-the-curve SKILL.mds (both mirrors), rewrites the
TTS invocation from ~/.claude/skills/media-use/... to the tracked
skills/hyperframes-media/scripts/heygen-tts.mjs, swaps npx hyperframes@latest
for the repo-local CLI in the gate step, and fixes a lint issue in Jake's
seam-gate.mjs (ternary-for-side-effect -> if/else).

Validated end-to-end by Home on Jun 23-29 (MP4 posted in C0ACCNHLG3U
thread 1784181166.041319). Independently reviewed R1/R2/R3 by Magi.

Co-authored-by: Jake Moran <jake@heygen.com>

.claude/skills/changelog-video/references/script-voice.md

@@ -0,0 +1,111 @@
+# Script + Voice: the two-layer contract
+
+The script is the single source of truth for BOTH the VO and the captions,
+written as token lines. The VO reads the `spoken` layer; captions render the
+`display` layer. This is a hard quality gate: a caption showing "jay-sawn" or
+a VO saying "juh-son" ("JSON" read literally) are both build failures.
+
+## Register (how it should sound)
+
+- Conversational, not release-notes. "The big one this week —" beats
+  "Theme 1:". Contractions welcome. Second person allowed ("your clips").
+- Informational, never salesy; no superlatives the changelog doesn't earn.
+- One breath per beat: sentences ≤ ~14 words; let punctuation pace the read.
+- Numbers with meaning stay ("fifteen releases"); commit hashes, PR numbers,
+  and version micro-detail are never spoken.
+- Open with the week + the marquee, close with the digest pointer
+  ("See everything at hyperframes dot heygen dot com").
+- **Teach the simple command.** When a feature has a one-line invocation (a
+  slash command, a CLI one-liner), the script says it verbatim ("start your
+  prompt with /figma…") and the mock shows it being typed — the command is
+  the visible CAUSE of the result. Slash commands speak as "slash <name>",
+  caption as `/name`.
+
+## Token-line format (`script-tokens.json`)
+
+```json
+{
+  "lines": [
+    {
+      "id": "l1",
+      "tokens": ["This", "week", "at", "HyperFrames,", { "display": "JSON", "spoken": "jay-sawn" }]
+    }
+  ]
+}
+```
+
+- A bare string = display and spoken identical.
+- An object = the layers diverge. `display` keeps standard spelling AND the
+  punctuation captions should show; `spoken` is what the TTS reads.
+- One line = one caption phrase (≤ ~40 chars of display text). Line grouping
+  is an authoring decision made here, not downstream.
+- Build `vo-spoken.txt` by joining every token's spoken form with spaces,
+  lines joined into sentences/paragraphs as punctuated.
+
+## Phonetics rules (ElevenLabs-style best practices, plain-text)
+
+HeyGen TTS takes plain text (no SSML), so pronunciation is controlled by
+spelling, hyphens, and spacing:
+
+1. **Initialisms** (each letter said): space or hyphen the letters —
+   `CLI → "C L I"`, `CDP → "C D P"`, `API → "A P I"`.
+2. **Acronyms said as words**: respell phonetically —
+   `JSON → "jay-sawn"`, `GSAP → "jee-sap"`.
+3. **Mixed / pronounceable compounds**: hyphenated LOWERCASE phonetics, one
+   fluid run — `ffmpeg → "ff-mpeg"` (ear-tested; the TTS reads "ff" as a
+   fluid "eff-eff"), `WebM → "web em"`, `OAuth → "oh-auth"`. Never spaced
+   capitals here: the TTS reads spaced caps as isolated letter names with
+   hard stops ("F F em-peg" comes out "eff… eff… em-peg"). Reserve spaced
+   capitals for TRUE initialisms (CLI, API) where a deliberate
+   letter-by-letter read is the goal. When candidates are close, generate
+   A/B takes of the real sentence and let the user pick by ear.
+4. **Versions/numbers**: expand — `v0.7.36 → "version zero point seven
+point thirty-six"` (usually: don't speak versions at all),
+   `1080×1080 → "ten-eighty by ten-eighty"`.
+5. **URLs**: `hyperframes.heygen.com → "hyperframes dot hey-jen dot com"`.
+6. **Filenames/extensions**: `.mp4 → "dot em pee four"` — or rephrase so the
+   extension isn't spoken.
+7. **Emphasis/pauses**: commas and em-dashes, never caps. Ellipses are
+   unreliable in TTS — use an em-dash.
+
+The shared vocabulary lives in `references/lexicon.json`
+(`display → spoken`). Consult it for EVERY technical term; if a term is
+missing, ask the user for the pronunciation and add the entry — never guess,
+never ship unheard. New entries: listen to that line in the generated VO
+before accepting.
+
+## Alignment (spoken timestamps → display captions)
+
+`heygen-tts.mjs --words` returns word timestamps of the SPOKEN text.
+`scripts/align-captions.mjs` walks the spoken stream against the token lines
+(one display token may cover several spoken words — "C L I" is three) and
+emits `captions.json`:
+
+```json
+{
+  "lines": [
+    {
+      "end": 3.1,
+      "w": [
+        ["This", 0.22],
+        ["week", 0.4],
+        ["JSON", 1.1]
+      ]
+    }
+  ]
+}
+```
+
+Each display word carries the start time of its FIRST spoken word; a line's
+`end` = the next line's start (last line: last word end + 0.6). The aligner
+warns `MISMATCH` when the heard word doesn't fuzzy-match the expected spoken
+form — every warning must be resolved (fix the lexicon spelling or the
+transcript) before the captions are trusted.
+
+## Caption rail (rendering)
+
+Per `captions-overlay`: a quiet OVERLAY, never a reserved band. One line,
+bottom-center (top: 1002px on 1080-square), TT Norms Pro 500 25px,
+ink .92, soft dark text-shadow, words fading in (0.12s) on their timestamps,
+phrase swaps as sets. Keep critical small text out of the bottom ~80px
+center span; everything else may run under the rail.