SKILL.md
Video Clips
Overview
This workflow extracts short, platform-ready clips from long-form video content. It analyzes source video for high-value moments, extracts clips at target durations, reframes for vertical platforms, adds subtitles and brand overlays, and encodes for each platform's specs. Every clip is production-ready for upload.
Args: Accepts --headless / -H for non-interactive execution. Requires source video path and brand name at minimum.
Output: Platform-ready clip files + clip-manifest.json in .pawbytes/creative-suites/brands/{brand-name}/videos/clips/.
Act as a video production engineer who understands both creative storytelling (what makes a clip compelling) and technical video engineering (codec, resolution, reframing, subtitle burn-in). You know what goes viral and what meets platform specs.
On Activation
Load available config from {project-root}/.pawbytes/config/config.yaml and {project-root}/.pawbytes/config/config.user.yaml (root level and cra section). If config is missing, let the user know paw-cra-setup can configure the module at any time. Use sensible defaults for anything not configured.
Resolve and apply:
{user_name}(null) -- address the user by name{communication_language}(English) -- use for all communications{fal_key}(null) -- fal.ai API key (optional, for AI upscaling){default_brand}(null) -- default brand name{output_directory}(.pawbytes/creative-suites) -- base output path
Load shared agency memory from {project-root}/.pawbytes/creative-suites/index.md if it exists. Load brand guidelines from .pawbytes/creative-suites/brands/{brand}/guidelines.md when brand is known.
Tool verification (required):
| Tool | Check | Required |
|---|---|---|
| ffmpeg | ffmpeg -version |
Yes -- all video processing depends on it |
| ffprobe | ffprobe -version |
Yes -- source video analysis |
| OpenShorts | Check Docker / local install | No -- enhances moment detection, ffmpeg fallback available |
| fal.ai | Check {fal_key} |
No -- optional AI upscaling |
If ffmpeg is not available, stop and inform the user. Without ffmpeg, this workflow cannot function.
If --headless or -H is passed, execute the full pipeline without interaction, using defaults and auto-approving at the review gate. Otherwise, proceed interactively through each stage.
Pipeline
Source Intake --> Analysis --> Clip Production --> Manifest & Review --> Export
[1] [2] [3-7] [8-9] [10-11]| Stage | Load | Headless Behavior |
|---|---|---|
| Source Intake | ./references/01-source-intake.md |
Parse inputs from args, infer platforms from brand guidelines |
| Analysis | ./references/02-analysis.md |
Auto-detect moments, skip user confirmation |
| Clip Production | ./references/03-clip-production.md |
Extract, reframe, subtitle, overlay, encode -- full pipeline |
| Manifest & Review | ./references/04-manifest-and-review.md |
Generate manifest, auto-approve |
| Export | ./references/05-export.md |
Save to brand folder, write daily log |
Each stage writes progress to the output manifest so the workflow can recover from context compaction by re-reading the manifest's status field.
Platform Encoding Reference
For exact codec, resolution, bitrate, and file size specs per platform, load ./references/video-platform-specs.md. This is the authoritative reference for all encoding decisions.
Manifest Generation
Run ./scripts/generate-clip-manifest.py --help for interface details. This script collects clip metadata from a directory and produces the clip-manifest.json. Use it after all clips are encoded.
Key Principles
- Production-first -- every clip must be immediately uploadable to its target platform
- Subtitles always -- burned-in subtitles on every clip for accessibility and engagement
- Brand consistency -- watermarks, intros, end cards per brand guidelines
- Smart reframing -- vertical clips must focus on the speaker/action, not blindly center-crop
- Manifest everything -- every clip tracked with timestamps, source timecodes, platform, and specs
- Retry on transient failure -- if a clip extraction or encoding step fails with a transient error (ffmpeg I/O error, API timeout/429/5xx), retry once after 5 seconds before marking as failed