konradre/video-production-skills

Fifteen agent skills, two host-side tools and a look library for producing video: short-form
ad campaigns, short films and music videos made from generated footage or from a client's own clips, and
narrated motion-graphics explainers drawn in code. The measurements behind the finishing rules are in the
references.
What is in the box
| directory | what |
|---|---|
| skills/video-production | the entry skill. Fixes the genre, rehydrates a paused project, routes the next step to the phase skill that owns it, and conforms supplied footage to one frame rate |
| skills/ad-spot-preprod · skills/film-preprod | pre-production for an ad campaign, polished or creator-style, or for a film and a music video. When the client supplies the picture, a curation procedure picks the shots from their pack of clips. For a new format, a reference teardown measures the shot lengths, face sizes, camera moves and speech timing of three to five real examples and turns them into a timed beat grid |
| skills/video-refs-continuity | the derived reference set, the start image, the continuity ledger and a refs gate that runs in code before any generation call |
| skills/video-prompt-dialects | the prompt compiled per venue dialect (Seedance 2.5 / 2.0, MiniMax H3, a web front end, two image models), with a linter, and the phone-native dialect a creator-style spot is prompted in |
| skills/video-gen-cost-gate | the venue table, the cost line and the operator's GO, the gated submit path, receipts at acceptance, detached polling. Five hosted routes sell the same Seedance 2.5 model, and the table ranks whichever of them you hold by marginal cost. The table is a roster you declare, and its rules shape a recommendation rather than refuse a run |
| skills/video-take-review | the per-seed read: continuity first, then the acceptance matrix, voids and usable windows, instruments with self-tests |
| skills/video-edit-edl | the edit as a derived EDL: the beat list from the script, a builder that computes every time, a script-fidelity gate |
| skills/spot-audio-assembly | voice-over, dubs, off-screen voices, sfx and music cues, room tone, captions, the loudness pass and its verification |
| skills/designed-elements | end cards, turntables, piece walls and drifts as deterministic HyperFrames compositions |
| skills/explainer-video | a narrated motion-graphics explainer drawn in code: sourced facts, a script whose spoken anchors time every scene, the voice and its word times, HyperFrames scenes and captions, a frame instrument and an audit of every on-screen string |
| skills/video-finish · skills/video-finish-qc | the per-clip finishing order (upscale, grade, grain, watermark, downscale, encode) and the spot pipeline that runs it from the EDL, plus the QC of the delivered file. A second finish, the phone-native tier, for a spot that has to read as phone-shot |
| skills/client-rounds | client notes located before anything moves, classified, executed by a ladder, delivered with a numbered ask |
| skills/mastering-audio | a standalone loudness master for a finished mix |
| tools/ | the local Topaz upscale wrapper and the unattended Resolve + Dehancer hero pass |
| look-library/ | five looks, each as a baked .cube and an authored Dehancer .drx, plus the cube-only colour tier of the phone-native finish, their YAML recipes and baker, the dial-ins, and a guide to the whole grading method |
| CHAIN.md | the handoff contract every skill runs under: a budget of three automatic hops, a visited set, stop conditions |
Each skill is a SKILL.md with the steps and their completion criteria, a references/ directory
with the contracts and measurements, and a scripts/ directory of small Python and shell tools.
Every Python and shell script prints its usage on --help and changes nothing, and the scripts that
work inside a project take its folder as --root .
skills/ is generated from a separate source tree and exported here, so an edit made to it in this
repository is overwritten by the next export. Read it, run it, copy from it; to change a skill, open an
issue rather than a patch.
What you need
The base install is an agent that reads SKILL.md — Claude Code and Codex both do — Python 3.10 or newer
with numpy, Pillow, scipy, PyYAML and OpenCV (pip install numpy pillow scipy pyyaml opencv-python), and
ffmpeg.
On Windows, run the kit inside WSL2. The skills give the agent bash commands, and WSL2 is where we built the kit.
AGENTS.md is the short orientation an agent reads before its first action; this file is the long form.