refactor(skills): rename graphic-overlays skill to talking-head-recut

[kiritowoo]

Rename the graphic-overlays workflow skill to talking-head-recut:

• move skills/graphic-overlays/ -> skills/talking-head-recut/
• update SKILL.md frontmatter name, H1, and self-references
• update all /graphic-overlays route references (hyperframes router, general-video, root + cli-template AGENTS.md/CLAUDE.md, docs, quickstart)
• update telemetry --skill flag, example composition id, timeline key
• update .prettierignore path and scripts/test-skills-fresh.sh

Identifier-only rename: the graphic-overlay card mechanism, design references, and trigger wording are unchanged.

What

Brief description of the change.

Why

Why is this change needed?

How

How was this implemented? Any notable design decisions?

Test plan

How was this tested?

• Unit tests added/updated
• Manual testing performed
• Documentation updated (if applicable)

[miga-heygen]

LGTM. Clean identifier-only rename across 37 files — no behavioral changes.

Verified: every + line in the diff uses talking-head-recut; zero new lines reference graphic-overlays. The rename covers all the surfaces: router skill table, workflow cheat-sheet, disambiguation bullets, workflow detail block, AGENTS.md/CLAUDE.md (root + cli-template), docs (prompting + quickstart), general-video cross-references, .prettierignore, test-skills-fresh.sh, NOTICE.md attribution, and the skill directory + SKILL.md frontmatter itself.

One formatting nit in docs/guides/prompting.mdx — the table alignment is slightly off after the rename (/talking-head-recut is longer than /graphic-overlays, so the | column shifts), but that's cosmetic and won't affect rendering.

CI still running. No issues.

Review by Miga

[james-russo-rames-d-jusso]

Verdict: 💚 Approve (posting as Comment per stamp protocol).

Reviewed at 25106f4c0f7de5a8b69bf1b87b8056d65f40990c. Layering on top of Miga's review — confirmed independently, plus a couple of notes Miga didn't surface.

---

Tiny identifier-only rename, 37 files / +24 / −24, 26 of those are pure file moves (assets + reference HTML). The risk surface is "did we miss a string somewhere" — verified end-to-end below.

💚 Verified — completeness of the rename

• git grep -nI 'graphic-overlays\|graphic_overlays\|graphicOverlays\|GraphicOverlays' at HEAD → 0 matches. main has 21. Total coverage.
• All caller surfaces touched: router skill table (skills/hyperframes/SKILL.md:61), workflow cheat-sheet (:72), workflow detail block (:118), root + cli-template AGENTS.md / CLAUDE.md, docs (prompting.mdx + quickstart.mdx), general-video/SKILL.md cross-ref (:18), .prettierignore, scripts/test-skills-fresh.sh (both the WORKFLOWS array :238 and the example echo :277), NOTICE attribution, frontmatter name:, H1, and the embedded HTML examples (data-composition-id, window.__timelines[...], --skill= invocation).

💚 Verified — no hidden coupling to the old slug

• The --skill= flag is regex-validated (SKILL_SLUG = /^[a-z0-9][a-z0-9-]{0,63}$/ at packages/cli/src/telemetry/skill.ts:14), not allowlist-gated. Both names match. render.ts:400 ignores invalid slugs with a warn-and-continue. No hardcoded KNOWN_SKILLS / VALID_SKILLS anywhere.
• Installation goes through upstream npx skills add heygen-com/hyperframes (packages/cli/src/commands/skills.ts), which resolves skills by directory + SKILL.md frontmatter name:. Both match (skills/talking-head-recut/ + name: talking-head-recut) → clean upstream resolution.
• Tests: zero *.test.* files reference either name; no vi.mock / fixture pin to the old slug; no export { X } from re-export gap.
• data-composition-id="talking-head-recut" and window.__timelines["talking-head-recut"] (SKILL.md :916, :1045, :1139) are local-scope to the composition's HTML (per docs/contributing/catalog.mdx:96 — id matches __timelines key per composition). Cosmetic edit, no renderer coupling.

🟡 Telemetry continuity gap (worth a heads-up, not a blocker)

• Per telemetry/skill.ts, the --skill slug is passed through to the event stream verbatim. Pre-merge events carry skill=graphic-overlays, post-merge events carry skill=talking-head-recut. Any PostHog / Datadog / dashboard query filtering on the old slug will silently see this as a new series rather than a continuation. If you have funnels / weekly cohorts on this skill, a quick skill IN ('graphic-overlays', 'talking-head-recut') patch on the affected dashboards (or a derived property in the analytics layer) keeps the timeline coherent across the cutover. Cheap fix; just noting it because nothing in the PR addresses it.

🟡 PR body has the unfilled template at the bottom

• The top of the description is excellent (clear scope + "identifier-only rename" framing). The bottom of ## What / ## Why / ## How / ## Test plan still has the placeholder text ("Brief description of the change", [ ] Unit tests added/updated, etc.). Won't block, but worth deleting the unfilled boilerplate before merging so the squash commit body is clean.

💭 Nit (echoing Miga)

• docs/guides/prompting.mdx:40 table-column alignment is slightly off after the rename — /talking-head-recut is 5 chars longer than /graphic-overlays, so the trailing | shifts. Cosmetic only, MDX renders fine.

CI: all completed required checks green (Lint / Format / Typecheck / Build / SDK contract+smoke / Studio: load smoke / Preview parity / Validate docs / regression / player-perf / CodeQL python+actions). A handful still in progress at review time (Test, CLI smoke (required), Windows render/tests, JS/TS CodeQL) — none failed. mergeStateStatus: BLOCKED is just waiting on the pending suite.

— Rames D Jusso

[miguel-heygen]

Verified at 25106f4. Clean identifier-only rename: branch-level grep shows no remaining old slug references, the skill directory/frontmatter/timeline/telemetry examples all use talking-head-recut, and all checks are green. LGTM.