---
name: video-interview-upload
description: Use this skill when processing a new Griot and Grits video interview for publication. Triggers when the user mentions uploading an interview, processing a new video, has a .mp4 or .mkv file to publish, wants to upload to YouTube for Griot and Grits, needs to generate tags or a YouTube description from an interview, or says things like "I have a new interview", "process this video", "upload to the channel", "new episode ready". Always use this skill for any Griot and Grits video publishing workflow.
---

# Video Interview Upload — Griot and Grits

Processes a new video interview from raw file to published YouTube video, including transcription, tag generation, and upload.

## What you'll need

- A `.mp4` or `.mkv` video file (or a `.txt` transcript if you already have one)
- YouTube OAuth2 credentials (see `references/setup_guide.md` if this is your first time)
- Python with required packages installed (see setup guide)

---

## Step 0 — Determine the working directory

First, check for a `.env` file. Look in both the current working directory and any path the user passed. If found, load it — it may define `GNG_VIDEO_DIR` and `GNG_YOUTUBE_CREDENTIALS`. A template for this file lives at `references/env.template`.

Then resolve the video processing folder using this priority order:

1. **Explicit path in the skill call** — if the user passed a folder path or file path (e.g., "process /home/shgriffi/videos/interview.mp4"), use that directory (or the file's parent directory).
2. **`GNG_VIDEO_DIR`** — from the `.env` file or the shell environment.
3. **Current working directory** — fall back to wherever Claude Code is running.

Resolve the YouTube credentials path:
- Use `GNG_YOUTUBE_CREDENTIALS` if set.
- Otherwise default to `<working_dir>/.youtube_credentials/client_secrets.json`.

Tell the user which directory and credentials path are being used before proceeding.

---

## Step 1 — Fetch the filter list

Before anything else, fetch the latest filters.yaml from GitHub:
```
https://raw.githubusercontent.com/griot-and-grits/gng-web/refs/heads/main/metadata/filters.yaml
```

This is the working copy for tag selection in Step 4. If the fetch succeeds, use it. If it fails (network unavailable), fall back to the bundled copy at `references/filters.yaml` and note that to the user so they know it may be slightly out of date.

---

## Step 2 — Locate the video and transcript

Look in the current directory for:
- A `.mp4` or `.mkv` file — this is the video to upload
- A `.txt` file with the same base name as the video (e.g., `interview.mp4` → `interview.txt`) — this is a pre-existing transcript

If the user mentions a specific filename, use that. If multiple video files exist, ask which one.

**If a `.txt` transcript file exists:** use it directly — skip Step 3.

**If no transcript exists:** proceed to Step 3.

---

## Step 3 — Transcribe the audio (if needed)

Run the transcription script:

```bash
python scripts/transcribe.py "<video_file_path>"
```

This extracts audio with ffmpeg and runs Whisper locally. It saves the transcript to `<video_basename>.txt` in the same directory. Tell the user it may take a few minutes depending on video length.

If the script fails because whisper or ffmpeg isn't installed, show the user this:
```
pip install openai-whisper
# ffmpeg: https://ffmpeg.org/download.html (or: brew install ffmpeg / apt install ffmpeg)
```

---

## Step 4 — Generate tags

Using the filter list fetched in Step 1, read the transcript carefully and select tags that genuinely apply — think about themes, locations, people, historical periods, and topics that a researcher or viewer might use to find this video.

**Rules:**
- **Use exact names only.** Every tag you select must be copied character-for-character from the YAML file — the `name` field under `tags` or `people`. Do not paraphrase, abbreviate, or invent similar-sounding alternatives. For example, if the list has `"Church Life"` and `"Faith"` as separate entries, do not combine them into `"Faith & Church"`. If the list has `"David E. Reid"`, do not shorten it to `"David Reid"`.
- Only include tags that clearly fit — don't pad the list
- For the `people` section: include a person only if they are actually featured/interviewed in the video **and their name appears in the people list**. If the speaker is unnamed or not on the list, do not add them.
- New tags: only suggest additions if a tag is so obviously essential that its absence is a clear gap. Set a high bar — if in doubt, don't add it. New tag suggestions are separate from the confirmed tag list and clearly marked for the user to decide.

Present the selected tags to the user and confirm before proceeding.

---

## Step 5 — Generate YouTube metadata

Based on the transcript, generate:

**Title:** Always starts with `"Griot and Grits - "`. Keep it short and compelling — 6 to 10 words after the prefix. Name the person if they're a featured subject. Reference the most striking or specific topic from the interview, not a generic summary.

Good: `Griot and Grits - Mrs. Clark Talks Black Prisoner Cadavers`
Too generic: `Griot and Grits - Interview About Medical History`

**Description:** One paragraph (4–6 sentences). Summarize who is speaking, what they discuss, why it matters, and what time period or location it covers. Write for someone discovering this video cold — give them enough to know if this story is relevant to their research or interests. Don't use bullet points.

**Tags:** Use the confirmed tags from Step 4 as the YouTube tags list.

Show the user the title, description, and tags together and ask for approval or changes before uploading.

---

## Step 6 — Upload to YouTube

Check whether the resolved credentials path from Step 0 exists.

---

### Option A — Credentials present (API upload)

Run the upload script:

```bash
python scripts/upload_to_youtube.py \
  --video "<video_file_path>" \
  --title "<approved_title>" \
  --description "<approved_description>" \
  --tags "<comma_separated_tags>" \
  --credentials "<resolved_credentials_path>"
```

The script will open a browser for OAuth2 on first run (or reuse the cached token), upload as **Public**, **not made for kids**, **Standard YouTube License**, and add the video to the **"Griot and Grits - Black Voices Worth Remembering, Black History Worth Sharing"** playlist. When complete it prints the YouTube video URL — share it with the user, then proceed to Step 7.

---

### Option B — No credentials (manual upload guide)

Tell the user credentials aren't set up yet, then walk them through uploading manually. Present the metadata in clearly labelled copy-paste blocks so they can work through YouTube Studio without switching windows:

---

**Go to:** https://studio.youtube.com → **Create** → **Upload videos** → select `<video_filename>`

While the file uploads, fill in the details:

**Title** (copy exactly):
```
<approved_title>
```

**Description** (copy exactly):
```
<approved_description>
```

**Tags** — in the Tags field, add each of these:
```
<tag_1>
<tag_2>
...
```

**Audience:** Select **"No, it's not made for kids"**

Click **More options**:
- **License:** Standard YouTube License
- **Category:** People & Blogs (or leave as-is)

Click **Next** through Checks, then on the **Visibility** screen:
- Set to **Public**
- Add to playlist: **"Griot and Grits - Black Voices Worth Remembering, Black History Worth Sharing"**

Click **Save**. Copy the video URL from the confirmation screen and share it here.

---

Then proceed to Step 7 for the remaining settings that apply to both upload methods.

---

## Step 7 — Final settings in YouTube Studio

After the upload is complete, open the video in YouTube Studio → **Details** tab → **Show more** and complete these settings (they cannot be set during upload):

- [ ] **Altered Content** → Check "My video contains altered or synthetic content (like AI-generated content)"
- [ ] **Remixing** → Ensure this is **OFF** (unchecked — do not allow others to remix)
- [ ] **Comments** → Set to **Hold all comments for review** (strict moderation)
- [ ] **Like count** → Check "Show how many viewers like this video"

Under **Video chapters**:
- [ ] Allow automatic chapters (should be on by default — confirm it is)

Under **Featured places** and **Automatic concepts**:
- [ ] Confirm both are enabled (YouTube enables these by default)

Show the user this checklist and remind them to complete it before the video goes live.

---

## Wrap-up

Once the user confirms the manual settings are done, summarize what was completed:

- Video uploaded: [YouTube URL]
- Tags applied: [list]
- Any new tags suggested for filters.yaml: [list, or "none"]

If new tags were suggested, ask the user whether they want to add them to filters.yaml now or later.