Skip to content

gooseworks-ai/goose-skills / skills/video/render-hook-replacement

render-hook-replacement

Verified

Replace an existing video's opening with a supplied clip or free kinetic text hook while retaining and verifying every original body frame, audio, captions and ending. Use for surgical hook tests rather than regenerating the complete ad.

FileKindSize
SKILL.mdskill md8.7 KB
config.example.jsonconfig254 B
references/hook-replacement.mdmarkdown4.9 KB
scripts/make_demo.pyscript2.9 KB
scripts/replace_hook.pyscript28.4 KB
scripts/replace_pack.pyscript3.6 KB
skill.meta.jsonconfig274 B
tests/smoke-test.mdmarkdown571 B
tests/test_replace_hook.pyscript11.9 KB

references/hook-replacement.md

View on GitHub
# Human version

A hook test changes the opening of an exact original version. It does not remake
the body. A shorter opening shortens the finished ad; a longer opening lengthens
it. Separate captions shift with that change, and burned captions stay untouched.

Supply a finished hook for free assembly, use a short free text card, or approve
a separately priced generation route. Watch the final complete video before saving
it as the chosen final.

---

# Agent version

## Freeze source before creating the opening

Bind the exact selected source/render identity, fetch its real bytes, and compute
SHA256. Record project/render IDs when the caller has them. Never replace that
source with the project's newest render, a template demo or a similar concept.
The renderer only handles local media; the integration adapter owns downloading,
access control, durable upload, saving and project/batch membership.

## Plan the cut

Watch the original opening and its transition into the body. Suggest an editable
boundary from the actual video, not a fixed three-second rule. Confirm how the new
opening leads into the first retained sentence/shot. Supply measured words when
available; a boundary inside a measured word blocks. A word list without a matching
source hash must not be reused across versions. Without measured words, the
manifest explicitly says speech was not measured and the listening review remains required.

Video is discrete frames. A boundary between frames keeps the first original frame
at/after that time and discloses the offset. Source audio is cut at the requested
time. Typical offset is under 33ms at 30fps. If this is unacceptable for a specific
join, ask for a frame-aligned boundary; do not silently alter the user's boundary.

## Choose only the needed route

| Treatment | Required new work | Paid generation |
|---|---|---|
| Supplied finished clip | Normalize hook dimensions/fps and assemble locally | None |
| Existing footage | Select/trim approved footage, then bind a finished hook | None, unless a separate paid service is chosen |
| Kinetic text | Approved short copy + licensed font; local fade/slide | None; Pillow draws the text |
| Voice over footage/text | New opening voice track using the chosen voice capability; local mux | Only the new voice track |
| Generated creator/video | New opening via chosen generation capability and usable references | Only the new opening calls |

Use published `create-vo-elevenlabs` for new narration and
`create-video-seedance-2-fal` for requested generated video after reading their
current entry points and quoting their current price. These capabilities are
optional, not requirements for supplied clips. Do not assume face/voice identity
matching without usable references. Do not auto-switch provider, add a narrator,
generate product images/music, or pay for retranscribing the entire original.
Retries which change paid behavior require an updated estimate and approval.

## Retain body and captions

The generic renderer uses a hard cut with no transition across retained body
frames. Browser-compatible High-profile libx264 CRF 12 retains original frame order/content
with every decoded frame checked at PSNR ≥45dB. Compressed bytes and some pixels
differ within this measured encode tolerance. Audio is re-encoded at AAC 256k with measured
sample-position error/correlation limits. No body gain, grading, crop, speed change
or new bed is applied. The exact original end card stays inside the retained body.

Separate SRT/JSON cues wholly before the cut are removed. Cues wholly after it shift
by `replacement_duration_sec - hook_end_sec`. Cues crossing the cut need original
measured words; only retained complete words remain. JSON per-cue styling fields
are preserved. ASS/VTT require a deliberate format adapter; do not strip styles by
pretending they are SRT. This operation does not burn a second caption layer.

## Review and failure

Read the manifest preservation results, then watch the actual complete output and
listen to the join. Decoded checks do not assess whether a new hook is persuasive,
brand-safe, readable or semantically connected. The review report must identify
the finished output hash and its frozen source. A failed/unreviewed result remains
a candidate and does not replace the chosen final or source file.

The renderer seeds `review.status: needs_review`, both SHA256 identities and
`checked_at: null`. Only after actual full-output viewing/audio/caption review,
set `review.status:'passed'`, keep `source_sha256` and `output_sha256` bound to
that manifest and record `checked_at` as an ISO datetime. Preserve the helper's
verification results and attach the actual passing quality report. A prior
generic report without these exact identities does not approve a new output.

For a pack, repeat against the **same frozen source** for every named opening.
Return one complete ad plus manifest per candidate. Project-version/batch saving
is a host decision handled by its integration adapter, never hidden in this script.
finishing-a-development-branchobra/superpowersUse when implementation is complete, all tests pass, and you need to decide how to integrate the workStars: 322kUpdated Claude CodeCodexVerifiedInstall
systematic-debuggingobra/superpowersUse when encountering any bug, test failure, or unexpected behavior, before proposing fixesStars: 322kUpdated Claude CodeCodexVerifiedInstall
test-driven-developmentobra/superpowersUse when implementing any feature or bugfix, before writing implementation codeStars: 322kUpdated Claude CodeCodexVerifiedInstall
skill-creatoranthropics/skillsCreate new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, edit, or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy.Stars: 198kUpdated Claude CodeCodexVerifiedInstall