lazyedit-publish-workflow

name	lazyedit-publish-workflow
description	Publish videos through LazyEdit, AutoPubMonitor, Nutstore AutoPublish, Shipinhao, YouTube, and Instagram from LabCanvas or WeChat worker tasks.

LazyEdit Publish Workflow

This repo-local skill mirrors the LazyEdit publish skill from /home/lachlan/DiskMech/Projects/lazyedit/references/skills/lazyedit-publish-workflow/SKILL.md and the active Codex skill. Use it when a WeChat/LabCanvas task asks to publish, re-publish, import, process, subtitle, or monitor a video. For LabCanvas-specific agent boundaries, also read references/lazyedit-agent-integration-handoff.md. The full LazyEdit-side handoff is /home/lachlan/DiskMech/Projects/lazyedit/references/AGENT_HANDOFF_LAZYEDIT_PUBLISH_2026_06_25.md.

Runtime Map

• LazyEdit repo/backend: /home/lachlan/DiskMech/Projects/lazyedit
• Studio app: http://127.0.0.1:18791/editor
• LazyEdit API: http://127.0.0.1:18787
• Publish CLI: scripts/lazyedit_publish.py
• AutoPubMonitor repo: /home/lachlan/DiskMech/Projects/autopub-monitor
• Nutstore import folder: /home/lachlan/Nutstore Files/AutoPublish/AutoPublish
• Remote AutoPublish host: ssh lachlan@lazyingart
• Remote publish API: http://lazyingart:8081/publish
• Remote tmux session: autopub

Core Rule

Prefer the LazyEdit CLI over manual browser work. It creates normal LazyEdit jobs, keeps the web queue in sync, and provides stable monitoring output. LazyEdit is the mature downstream video tool: do not rebuild subtitle correction, translation, metadata, logo/subtitle burn, browser-safe packaging, or platform posting in LabCanvas. LabCanvas agents prepare exact source video evidence plus context files, call LazyEdit, monitor terminal evidence, and send artifacts/status back to WeChat.

cd /home/lachlan/DiskMech/Projects/lazyedit
source ~/miniconda3/etc/profile.d/conda.sh
conda activate lazyedit

Safety Rules

• Use --no-publish before a real publish when testing packaging, subtitles, or logo output.
• Publish exactly once after the final MP4/ZIP is correct. When the user explicitly asks to publish, --no-publish is only a temporary quality gate; continue to the real publish automatically unless login, CAPTCHA, payment, consent, or another manual block prevents it.
• Real publishes should use polished subtitles and the configured LazyEdit logo unless the user explicitly says otherwise.
• Verify logo settings before publish:

curl -fsS http://127.0.0.1:18787/api/ui-settings/logo_settings | jq .

• Normal logo outputs end in _subtitles_logo.mp4.
• Use --correction-prompt-file for full transcript/story context and --metadata-prompt-file for a short public-facing brief. Do not pass a full script as metadata context.
• Generated-video LazyEdit runs must prefer the worker-created lazyedit_correction_context.md and lazyedit_metadata_brief.md from task.preflight.lazyedit_context; story/prompt files from the browser monitor are fallback only.
• The correction context must include the WeChat message sent with the video. For AI-generated video publication, append the generated story/script and Xiaoyunque/Seedance prompt before calling LazyEdit.
• Use the resumed Codex/AgInTi worker agent to call routines, scripts, and CLI commands. Deterministic code may isolate sources, guard duplicates, probe status, verify terminal evidence, and enforce artifact delivery, but it must not become a separate hardcoded publishing workflow.
• Silent or nearly silent videos may produce empty transcripts and burn=skipped. This is acceptable when transcribe/translate/caption/keyframes are complete; continue metadata generation, cover extraction, publish queue submission, and terminal platform verification instead of waiting forever or swapping in an older video.
• AutoPublish browser uploads require a web-safe MP4 inside the ZIP. LazyEdit must package _highlighted.mp4 as H.264/AVC (avc1), yuv420p, AAC audio, and +faststart. If the selected source/burn output is HEVC/H.265, AV1, or another browser-risk codec, transcode it during publish-bundle preparation before sending the ZIP.
• Do not commit generated ZIPs, runtime media, temporary prompts, cookies, or queue snapshots.

WeChat Video Import

For a video shared in a monitored WeChat group or DM, first copy the same-chat source into the Nutstore AutoPublish watcher:

PYTHONPATH=src python -m agenticapp wechat autopublish-video \
  --chat "<chat>" \
  --message-local-id VIDEO_LOCAL_ID \
  --sync \
  --fetch-gui \
  --since-minutes 720 \
  --json

This opens the isolated WeChat desktop when needed, clicks the latest visible video so WeChat caches the MP4, syncs media, and atomically writes a *_COMPLETED file into the Nutstore watcher. Use --list --json to inspect candidates and --source /abs/video.mp4 when the exact source file is known. When a task references a specific WeChat video row, always pass --message-local-id; this prevents a nearby older cached MP4 from being copied or published by mistake. If the exact row cannot be cached, fail closed and ask for a resend or a GUI cache retry rather than falling back to another video.

After import, find the LazyEdit video_id:

tmux capture-pane -pt autopub-monitor:0.1 -S -100 | tail -n 100
tmux capture-pane -pt autopub-monitor:0.2 -S -100 | tail -n 100
curl -fsS http://127.0.0.1:18787/api/videos | jq '.videos[:20] | map({id,title,created_at,file_path})'

WeChat Context To Subtitle Correction

When the publish request comes from WeChat, preserve the full source-scoped task packet generated by the monitor:

• current coalesced request;
• quoted message text;
• recent same-chat history;
• source/reference local_id rows;
• visible media metadata such as title, filename, URL, MD5, and video row;
• synced file paths from .private/downloads/<chat>/....

Write that context under the worker artifact directory, for example lazyedit_correction_context.md, and pass it as --correction-prompt-file. Use it to fix names, terminology, and obvious ASR mistakes. Do not invent lines that are unsupported by the audio/video or chat context.

Create a separate short file, for example lazyedit_metadata_brief.md, for viewer-facing title, description, keywords, and platform notes. Pass that file as --metadata-prompt-file. Never reuse the full chat history or full script as metadata context.

Process Then Publish

Use separate subtitle-correction and metadata context files:

python scripts/lazyedit_publish.py \
  --video-id VIDEO_ID \
  --use-current-settings \
  --correction-prompt-file /abs/full_context.md \
  --metadata-prompt-file /abs/metadata_brief.md \
  --no-correct-subtitles \
  --steps keyframes,caption,transcribe,polish,translate,burn,metadata_zh,metadata_en,cover \
  --platforms shipinhao,youtube,instagram \
  --guided-monitor \
  --remote-log-command "ssh lachlan@lazyingart 'tmux capture-pane -pt autopub:0 -S -140 | tail -n 140'" \
  --wait \
  --poll-seconds 10 \
  --process-timeout 3600 \
  --publish-timeout 7200

For direct upload from a local MP4:

python scripts/lazyedit_publish.py \
  --video /abs/video.mp4 \
  --title TITLE_COMPLETED \
  --use-current-settings \
  --correction-prompt-file /abs/full_context.md \
  --metadata-prompt-file /abs/metadata_brief.md \
  --correct-subtitles \
  --correction-source polished \
  --platforms shipinhao,youtube,instagram \
  --wait \
  --poll-seconds 10

Publish Existing Output

Use --no-process when the user says "same output", "last run", or the final processed video already exists:

python scripts/lazyedit_publish.py \
  --video-id VIDEO_ID \
  --use-current-settings \
  --platforms shipinhao,youtube,instagram \
  --no-process \
  --wait \
  --poll-seconds 10

Platform-only variants:

python scripts/lazyedit_publish.py --video-id VIDEO_ID --use-current-settings --platforms shipinhao --no-process --wait --poll-seconds 10
python scripts/lazyedit_publish.py --video-id VIDEO_ID --use-current-settings --platforms youtube,instagram --no-process --wait --poll-seconds 10

Context Files

Subtitle correction context may include the full story, transcript, prompt, or user notes. Treat it as evidence, not a verbatim script: fix obvious ASR errors, names, objects, and broken phrases without inventing unsupported dialogue.

Metadata context should be short and public-facing:

• one hook sentence
• characters and setting
• central conflict, joke, or emotion
• desired title tone
• 8 to 15 keywords or hashtags
• instruction not to reveal every beat or line

Manual Subtitle Quality Pass

Before real publish, inspect polished subtitles when precise context exists:

sed -n '1,180p' DATA/VIDEO_FOLDER/*_mixed_polished.md
rg -n "bad term|broken term|ASR artifact" DATA/VIDEO_FOLDER/*_mixed_polished.*

Keep .json, .srt, and .md aligned if hand-editing. Prefer LazyEdit save endpoints when available; use database recovery only for clear duplicate-worker status corruption.

Monitoring

Local LazyEdit queue:

curl -fsS http://127.0.0.1:18787/api/autopublish/queue | jq '.jobs[:8] | map({id,video_id,status,platforms,remote_status,remote_job_id,error})'

Remote AutoPublish queue:

curl -fsS http://lazyingart:8081/publish/queue | jq '.jobs[:8] | map({id,status,platforms,filename,error,updated_at})'

Remote browser automation:

ssh lachlan@lazyingart 'tmux capture-pane -pt autopub:0 -S -120 | tail -n 120'

AutoPubMonitor panes:

tmux capture-pane -pt autopub-monitor:0.0 -S -120 | tail -n 120
tmux capture-pane -pt autopub-monitor:0.1 -S -120 | tail -n 120
tmux capture-pane -pt autopub-monitor:0.2 -S -120 | tail -n 120
tmux capture-pane -pt autopub-monitor:0.3 -S -120 | tail -n 120

Shipinhao Notes

• Shipinhao may require WeChat QR or email login. Keep the long wait running after the user scans.
• Expected success log includes Successfully published on ShiPinHao.
• Do not bypass CAPTCHA, login, or consent pages. Open the isolated browser for human assistance and ask the user to approve continuation.

Recovery Checks

If processing appears stuck, inspect processes before killing anything:

ps -eo pid,ppid,cmd | rg 'vad_lang_subtitle|HandBrakeCLI|scripts/lazyedit_publish.py'

Only stop a duplicate worker when it is clearly redundant and a valid completed output exists. Then confirm queue state before publishing.

Handoff Report

Final status back to WeChat or the terminal should include:

• LazyEdit job id
• remote job id, if any
• platforms
• current status or blocking human step
• whether processing was reused or rerun
• output MP4/ZIP paths only when safe to share