raw-video-clip-intake-ops

name	raw-video-clip-intake-ops
description	Sort raw camera-card clips into the right short-form video project folders. Use when a creator says they filmed several videos on a DJI/camera card or external drive and needs the clips grouped by topic, separator shots, timestamp order, or spoken context into weekly batch folders while preserving raw sidecars such as MP4, WAV, LRF, SRT, or proxy files and writing intake manifests before rough-cut editing.
memory_tags	["domain:social-video","workflow:raw-clip-intake","skill_role:operator","repo_boundary:tools","inputs:camera-card","outputs:intake-manifest","risk:medium"]

Raw Video Clip Intake Ops

Purpose

Turn a messy camera card into project-owned raw folders without losing sidecars or making future editing agents rediscover what each take was.

Use this before shortform-rough-cut-ops. This skill answers: "Which raw files belong to which video project?" Rough-cut skills answer: "What should survive inside those clips?"

Core Rule

Do not delete or rename source-card files during intake. Copy or hard-link raw sidecar sets into project folders, then write an INTAKE.md manifest with the source path, grouping rationale, clip list, durations, and uncertainty.

Workflow

1. Resolve context.
   • Identify the source card/folder.
   • Identify the destination weekly batch folder.
   • Identify the intended video topics and their rough filming order.
   • Load the relevant weekly video plan or private video-operator ledger when available, because project slugs may already exist.
2. Inventory the source.
   • List candidate video files by timestamp and filename order.
   • Preserve sidecar relationships by shared stem, such as .MP4, .WAV, .LRF, .SRT, and proxy files.
   • Use ffprobe for duration/resolution when available.
   • Build a contact sheet when visual grouping matters.
3. Infer groups.
   • Use filename order and capture timestamps as the first pass.
   • Use user-stated topic order as a strong signal.
   • Use short clips, camera-bag shots, phone/prop shots, repeated setup shots, and long gaps as separators.
   • Treat long talking-head clips as likely main takes; treat very short clips as setup, false starts, separators, or pickups until confirmed.
   • If two adjacent groups are ambiguous, stop and ask for confirmation before copying, or copy into a clearly labeled needs-review folder.
4. Create destination folders.

   • Use the existing local batch shape when present:

     YYYY-MM-DD_video-slug/
       raw/
       assets/
       editor-projects/
       exports/
       final-videos/
     

   • Do not hard-code local media roots in the reusable skill. Resolve them from user input or private repo config.

5. Copy sidecar sets.
   • Copy every file sharing the selected clip stem into that project's raw/ folder.
   • Prefer non-destructive copy by default. Use hard links only when the user asks and the source/destination are on the same volume.
   • Run a dry-run first when the grouping was inferred rather than explicitly supplied.
6. Write manifests.
   • Add INTAKE.md to each project root.
   • Include source folder, destination, copy mode, original files left in place, clip durations, separator notes, and unresolved questions.
7. Update private state.
   • Record created project folders in the weekly plan or video-operator ledger.
   • Capture workflow friction that should improve the next version of the skill, such as slow external-drive copies or missing labels.

Script

Use scripts/intake_camera_card.py for deterministic inventory, dry-runs, copying, and manifest generation.

Example grouping plan:

{
  "groups": [
    {
      "slug": "2026-07-08_story-time-example",
      "title": "Story Time Example",
      "clip_ids": ["0008", "0009", "0010"],
      "notes": {
        "0008": "short lead-in",
        "0009": "false start or pickup",
        "0010": "main take"
      }
    }
  ]
}

Dry-run:

python scripts/intake_camera_card.py \
  --source-dir "<camera-card>/DCIM/DJI_001" \
  --batch-dir "<media-root>/batches/YYYY/YYYY-Www_video-batch" \
  --groups-json grouping-plan.json \
  --dry-run

Apply:

python scripts/intake_camera_card.py \
  --source-dir "<camera-card>/DCIM/DJI_001" \
  --batch-dir "<media-root>/batches/YYYY/YYYY-Www_video-batch" \
  --groups-json grouping-plan.json \
  --apply

Generate a contact sheet during inventory:

python scripts/intake_camera_card.py \
  --source-dir "<camera-card>/DCIM/DJI_001" \
  --inventory \
  --contact-sheet /tmp/camera-card-contact-sheet.jpg

Output Contract

Return:

• source folder inspected
• destination batch folder
• grouping table: project slug, clip IDs, file count, size
• created or reused project folders
• manifest paths
• unresolved grouping uncertainty
• next editing step, usually transcription-first A-cut with shortform-rough-cut-ops

Guardrails

• Never delete source-card files.
• Never copy private raw footage into a public git repo.
• Do not rename raw clip files during intake unless collision handling requires a suffix.
• Do not silently drop sidecars.
• Do not treat a camera-bag separator as proof by itself; combine it with timestamp order, user topic order, and clip duration.
• If a source file is still being written or a copy leaves temporary files, stop and verify before continuing.