AI agent skill · by @antoniolg

Short Publish

End-to-end workflow for turning a local video into transcripts, burned subtitles, and scheduled multi-network posts via PostFlow CLI. Use when given a video path and publication date/time to transcribe, create copy for LinkedIn/X/IG/YouTube, upload the subtitled MP4, and schedule the content with `postflow`.

Archived 2026-10-23 · Source updated 2026-05-18. Check the original repository for the latest version.

SKILL.md instructions

# Short Publish

## Overview

This skill automates the complete "video → subtitles → PostFlow" pipeline: run Whisper-based transcription, burn subtitles with the bundled Python script, turn the transcript into a multi-platform copy block, and schedule social posts through the PostFlow CLI.

## Inputs & Prerequisites

- **Arguments:**
  - `PATH` – absolute path to the source video (MOV/MP4/etc.).
  - `DATETIME` – publication date/time (accepts natural language like "tomorrow 09:00"). Use `date` to confirm the current timestamp if needed.
- **Tooling:** use the `postflow` CLI (`postflow media upload`, `postflow posts create`) and refer to [`postflow-cli`](../postflow-cli/SKILL.md) for command details.
- **Script dependency:** `scripts/transcribe_burn.py` wraps Whisper, ffmpeg, and auto-gain. Requires Python 3.8+, ffmpeg, and `openai-whisper` installed for the user; no extra configuration is needed inside this skill.
- **Timezone:** default to Europe/Madrid. In winter assume UTC+01:00 (CET) when presenting final schedules if the `date` command does not provide the offset.

## Workflow

1. **Collect inputs**
   - Confirm the provided `PATH` exists; stop with a descriptive error if not.
   - Resolve `DATETIME` to an ISO timestamp. Use `date -j -f` or another deterministic macOS command when the input is natural language so PostFlow receives an unambiguous value.

2. **Transcribe and burn subtitles**
   - Run the bundled helper: `python3 scripts/transcribe_burn.py "$PATH"`.
   - Outputs (all written next to the original video):
     - `<stem>.srt`, `<stem>.ass`, `<stem>.txt`, `<stem>_caption.txt`, `<stem>_subtitled.mp4`.
   - The `_subtitled.mp4` is the media you will upload; everything else is transient reference material. Remove the generated artifacts (srt/ass/txt/caption/mp4_subtitled/normalized wav) once they have been read and the upload succeeds—never delete the original video.

3. **Generate the social copy**
   - Read `<stem>.txt` for the full transcript.
   - Apply the exact copywriting prompt below to the transcript; do not improvise structure or tone beyond the template.

     ```
     Act as an expert LinkedIn copywriter building authority content.
     Transform the TRANSCRIPT into a case-study or practical-lesson post with this structure:
     1. Hook headline with a leading emoji.
     2. 2-3 sentence context introducing the situation.
     3. Structured core (use 1️⃣/2️⃣/3️⃣ or ✅ and bold keywords per line).
     4. Closing takeaway line.
     5. Optional P.S. only when the transcript mentions an offer/event.

     Style rules: short paragraphs (1-2 lines), intentional emoji usage, no invented facts, stay faithful to the transcript.
     ```

   - Reuse the single output block verbatim for LinkedIn, X, and Instagram, and as the YouTube description (light line breaks allowed). Craft a YouTube title ≤100 characters from the same content.

4. **Upload the subtitled video**
   - Upload the `_subtitled.mp4` and capture `media_id`:
     ```bash
     postflow --json media upload --file "<stem>_subtitled.mp4" --kind video
     ```
   - Use the returned `id` as `media_id` for all posts.

5. **Schedule posts via `postflow posts create`**
   - Accounts come from `~/.config/skills/config.json` under `postflow.groups.short_publish` and `postflow.accounts`.
   - For X aliases, use `postflow.defaults.x` (default `x-es`) unless the user explicitly requests another account.
   - For each account, create a scheduled post with the same copy block and the uploaded media:
     ```bash
     postflow posts create \
       --account-id <acc_id> \
       --text "<copy_block>" \
       --media-id <media_id> \
       --scheduled-at <ISO8601>
     ```
   - If a thread is needed, use `--segments-json` instead of `--text`.

6. **Report completion**
   - Confirm each scheduled post by echoing returned IDs and scheduled time in CET (UTC+01:00 during winter). Example: `LinkedIn pst_... → 2025-01-11T10:00:00+01:00 (CET)`.

## Resources

- `scripts/transcribe_burn.py`: Whisper + ffmpeg pipeline used in Step 2. Copy-safe to reuse elsewhere but do not edit unless the video workflow changes. Running the script produces all intermediate assets and the burned MP4 referenced throughout the workflow.