agentleFS
beratmemisogluberatmemisoglu/claude-viral-animasyon-sablonu/skills/daily-film-pipeline/SKILL.md

daily-film-pipeline

Skill4 starsChanged 15 days agoMITView on GitHub

The once-a-day autonomous production run of the YouTube channel "The Other Self" — preflight checks, competitor-radar → today's original idea, inner-voice-film → screenplay + Remotion animation + ElevenLabs twin voice + calm score + QA + YouTube render, other-self-thumbnail → 3 concepts on paper, the agent picks the CTR winner and generates only that ONE thumbnail (Higgsfield), metadata, YouTube Data API upload + thumbnail + publish, logs and a Turkish Discord report. Use it for the daily cron job and whenever the owner says "bugünün filmini yap", "günlük üretim", "daily run", "otomatik film".

  • Reads credentials

The file

We split this skill into the rules it gives an agent

2 rules. About 0 are common, 0 are distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈0

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Run the full competitor-radar procedure → project/films/<slug>/BRIEF.md (slug = lowercase title word).

    line 23
  2. Very specific:

    Write project/films/<slug>/SCREENPLAY.md from BRIEF.md: 13 beats, frame ranges summing to 4950 (2:45), voice ledger, couplets, slogan, …

    line 26

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Daily Film Pipeline — The Other Self

One run = one finished, uploaded film. Work in ~/the-other-self. Follow .hermes.md rules. The owner decided (2026-10-15) that this runs every day without a screenplay approval gate; quality gates below replace it.

0. Preflight (stop and report if anything fails)

bash ~/.claude/skills/daily-film-pipeline/scripts/preflight.sh — checks: already published today? (one film per day), ELEVENLABS_API_KEY set, YouTube token valid, disk ≥ 5 GB, studio typecheck. Then call mcp__higgsfield__balance (need ≥ 15 credits). On failure: do NOT continue; report the exact problem in Turkish.

1. Idea (skill competitor-radar)

Run the full competitor-radar procedure → project/films/<slug>/BRIEF.md (slug = lowercase title word).

2. Screenplay (skill inner-voice-film, FILM mode steps 1–3)

Write project/films/<slug>/SCREENPLAY.md from BRIEF.md: 13 beats, frame ranges summing to 4950 (2:45), voice ledger, couplets, slogan, PACKAGE. Run the skill's QA checklist (stamp < 0:15, rock bottom 31–40 %, every voice line paid off, no competitor lines, no red). Fix before building. No approval wait.

3. Build (inner-voice-film steps 4–6)

  • cp -r studio/src/films/procrastination studio/src/films/<slug> as the template (newest, most complete); rewrite timeline.json, scenes, Film.tsx names, and register the film + per-scene compositions in studio/src/Root.tsx (copy the procrastination Folder block). Reuse src/lib/ (rig, stage, text, props, TitleCard, Twin, Bedroom-like sets). New props go into the film folder.
  • cd studio && npx tsc --noEmit must pass.
  • Visual QA (mandatory): ./scripts/stills.sh <slug> "<scene:frames;…>" out/qa/<slug>-N.jpg for every scene (2–3 frames each), open every sheet with the vision tool, fix: text overlapping text/heads/windows/DAY counter, off-model character (round head + sprout, oval eyes), limbs detached, empty frames, text outside letterbox. Repeat until clean (max 4 rounds).

4. Voice + music

  • src/films/<slug>/voice-lines.json = every blue line (n, text, style whisper|taunt|boss, spoken for numbers).
  • Copy studio/scripts/procrastination-cues.mjs → <slug>-cues.mjs, change the scene ids, voice placements ([S.<scene> + <frame of the blue text>, n, gain]) and SFX for the new sets. Keep the calm music rules.
  • ./scripts/audio.sh <slug> → loudness must print about −15 LUFS.

5. Render

./scripts/render-youtube.sh <slug> → project/films/<slug>/<SLUG>-youtube.mp4. Verify with ffprobe: 1920×1080, duration ≈ 165 s, audio present. Pull 6 frames with ffmpeg at the milestones (0:08, 0:45, 0:55, 1:05, 2:00, 2:35) and look at them.

6. Thumbnail — ONE final image (owner decision 2026-10-15; no A/B test on the channel)

  1. Hook Engine (skill other-self-thumbnail §6b): write 3 concepts on paper (H1 absurd · H2 twin sabotage · H4 extreme emotion), each with text (1–3 words, "." or "?"), scene, emotion, light, title pairing.
  2. Pick the CTR winner yourself: score each with the 6-question stop-scroll score (0–2 each) + 2 extra points: (a) differs from the competitor thumbnails you saw today, (b) text and title do different jobs. Use project/LEARNINGS.md (what got clicks before) to break ties. Write the 3 scores + the reason into YOUTUBE.md.
  3. Generate only the winner in Higgsfield (nano_banana_pro, reference project/brand/profile/ref-hero-standing.png, 2 candidates in one call, ≈4 credits). Pick the on-model candidate (round bald sand head, ONE leaf sprout, no red, no gibberish text). If both drift, regenerate once (max ≈8 credits per film); if still bad, use the runner-up concept once. Never exceed ≈8 credits.
  4. Overlay the text with generator/render.py (baseImage file:///home/ubuntu/...) → project/films/<slug>/thumbs/<SLUG>-thumb.png
    • _320 preview; open the 320 px preview with vision and confirm the word, the face and the glow read at phone size.
  5. This single image is the one uploaded with the video.

7. Metadata + upload

  • project/films/<slug>/YOUTUBE.md (title, 2 alts, description with REAL chapter timestamps from timeline.json, 3 hashtags, tags, pinned comment, Shorts ideas).
  • project/films/<slug>/description.txt and upload.json: {"file": "project/films/<slug>/<SLUG>-youtube.mp4", "thumbnail": "project/films/<slug>/thumbs/<SLUG>-thumb.png", "title": "<SLUG> | An Animated Short Film", "description_file": "project/films/<slug>/description.txt", "tags": [...], "privacy": "private", "publish_at": "<output of: python3 ~/.claude/skills/daily-film-pipeline/scripts/publish_time.py>", "category": "1", "synthetic": true}
  • .venv-yt/bin/python studio/scripts/youtube/youtube_upload.py --meta project/films/<slug>/upload.json (run from ~/the-other-self). It prints VIDEO id + privacy. NOTE: while the Google API project is unaudited, YouTube forces uploads to stay private and ignores publish_at → then tell the owner to press "Publish" in YouTube Studio at 17:00 UK time.
  • Publish time = every day 17:00 Europe/London (owner decision 2026-10-15; DST-safe via publish_time.py). The cron job starts at 07:00 UTC so the film is ready hours before. If production runs late, publish_time.py falls back to "now + 20 min" and the report must say so.

8. Log + report

  • Append to project/films/PUBLISHED.md (date, film, id, URL, privacy, notes); mark the topic [x] in TOPICS.md.
  • Discord report (Turkish, short): film + link, what was verified (stills rounds, loudness, duration), thumbnails used (the winner + why it beat the other 2 concepts), Higgsfield credits spent, today's competitor insight, anything the owner must do (e.g. press Publish).

Failure policy

Any step that still fails after 2 honest fix attempts → stop, do not upload, keep the files, report the error and the step. Never upload a film that was not visually checked. Never skip the voice (if ElevenLabs fails, stop).

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.