agentleFS
graphed-incgraphed-inc/marketing-skills/skills/post-to-video/SKILL.md

post-to-video

Skill73 starsChanged 4 days agoMITView on GitHub

Turn an X or LinkedIn post into an editable Tesseract motion-graphic video, 1080x1350 MP4, timed to a music beat grid. The visual concept comes from mining a shell-generated random string for hidden patterns (never shown on screen). Use when the user pastes a post and asks for a motion graphic or video for it, or /graphed:post-to-video.

The file

We split this skill into the rules it gives an agent

8 rules. About 0 are common, 3 are distinctive, and 5 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 5

    Very specific

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

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

    Use this skill's own directory.

    line 19
  2. Very specific:

    If setup.sh says tsrct is missing or the wrong version, follow tesseract-motion/references/installation.md (the skill installed beside this …

    line 27
  3. Distinctive:

    If the user attaches a screenshot of real data, redraw it natively (read values off the pixels), state which values are exact vs.

    line 37
  4. Distinctive:

    Look past the surface: hidden words/abbreviations, palindromes, a digit sum that also appears as a substring, clusters, the length's …

    line 46
  5. Very specific:

    Check references/catalog.md and pick a system that hasn't been used.

    line 53
  6. Distinctive:

    Default 1080x1350, light theme, one accent.

    line 86
  7. Very specific:

    Write videos/<slug>/.tesseract-work/NOTES.md: brief, seed patterns → concept, look, music + in-point

    line 107
  8. Very specific:

    Add a row to references/catalog.md (and music.md "Used by").

    line 109

≈ “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

post-to-video

One post in, one finished video out: videos/<slug>/<Name>.mp4 plus the editable <Name>.tsrct, a build.py that generates every layer and keyframe, and a NOTES.md.

The pipeline is Python → editable JSON → tsrct project commit → tsrct export. All shared code lives in lib/motion.py. Each video only writes scenes. This skill sits on top of tesseract-motion (CLI install, schema, filmstrip review): read that skill's references only when you need something the library doesn't cover.

0. Setup (once per machine)

Use this skill's own directory. When the skill loads, that directory is the "Base directory for this skill". Do not assume .claude/skills/post-to-video; a registry install lives at skills/post-to-video inside the Graphed plugin.

bash "<skill-dir>/scripts/setup.sh"    # checks pinned tsrct, creates the Python venv, checks Linux libraries
source "<skill-dir>/scripts/env.sh"    # sets $PY, $T, $POST_TO_VIDEO
"$T" --version                         # must match tesseract-motion/references/cli-version.txt

If setup.sh says tsrct is missing or the wrong version, follow tesseract-motion/references/installation.md (the skill installed beside this one). That guide downloads the pinned CLI zip, checks its checksum, and runs the official install.sh. Then run setup.sh again.

setup.sh does not download tsrct and does not use sudo. On Linux it also checks for a Vulkan driver and for ffmpeg with libx264. If it prints an install command, run that command, then run setup.sh again. Later steps assume env.sh has been sourced.

1. Read the post

  • Pull out the beats of the argument: hook → mechanism steps → proof → CTA. One scene each; 5–8 scenes, 25–35 s total.
  • Pacing: hook ≈ 10 beats, most scenes 8–12 beats, one "hero" scene up to ~18. List items land every 1–2 beats, not 4. Staggered grids/tiles 40–110 ms apart. Hold a finished state ~1–1.5 s, then cut. No idle beats waiting for the next element. A cut near 48 s is too slow; about 28 s is the pace that works. Stay in 25–35 s.
  • Flag, don't silently fix: likely typos in product or model names, numbers that don't match supplied data, and claims you can't support. Use the sensible reading on screen and list every change in the final message.
  • Graphed CTA (logo + get started free → graphed.com/platform) only if the post mentions Graphed.
  • If the user attaches a screenshot of real data, redraw it natively (read values off the pixels), state which values are exact vs. read off the chart, and keep the post's claim and the chart consistent or flag the gap.
  • If the user names a design reference (URL or screenshot), that overrides the seed concept for the look. Read it (ask before opening a browser; Cloudflare sites block curl/WebFetch), match its type, colour, components and motion, and keep the Graphed mark. Never put a third-party logo on screen.

2. Seed → concept

bash $POST_TO_VIDEO/scripts/seed.sh videos/<slug>-seed.txt 120    # prints palindromes, runs, doubles, hidden words, digit sums, factors

Look past the surface: hidden words/abbreviations, palindromes, a digit sum that also appears as a substring, clusters, the length's factors (96 = 8×12 → a 12-column grid). Tie 2–4 findings to the post's message and turn them into ONE visual system (a world + a signature motion + one persistent element). Examples that worked: "007" → version numbers → a changelog; "p2"+"hde" → hide on page 2 → a scrolling page with a rank ruler; "vOv" → a face → the bot avatar.

  • Never show the string, or a recognisable chunk of it, on screen. Save it in .tesseract-work/seed.txt.
  • Check references/catalog.md and pick a system that hasn't been used. Different from the last video, not a reskin.
  • Tell the user in 2–4 lines which patterns you found and what they became.

3. Music

$PY $POST_TO_VIDEO/scripts/beats.py $POST_TO_VIDEO/assets/music/<track>.mp3 --need 50
$PY $POST_TO_VIDEO/scripts/beats.py $POST_TO_VIDEO/assets/music/<track>.mp3 --zoom 31.2    # confirm a drop

Choose from references/music.md (prefer an unused track/section). Put the quiet part under the first line of the hook and the drop on beat 3–4 (≈2 s) under the key line. All scene boundaries are beats. Weak-grid tracks (strength < 1.25) are out.

4. Build

bash $POST_TO_VIDEO/scripts/new.sh <slug> <CamelName> <track>     # claims videos/<slug> (fails if it exists), copies fonts/sfx/logo/music, writes template build.py + rebuild.sh
cp videos/<slug>-seed.txt videos/<slug>/.tesseract-work/seed.txt && rm videos/<slug>-seed.txt

Edit videos/<slug>/.tesseract-work/build.py (the template shows hook / list / counter / CTA):

  • setup(ROOT) then palette, B = Beats(period, [scene beats…]), MUSIC_SRC0 = <in-point ms>.
  • Scene groups: group(name, S[i], D[i], layers) + enter_scene / exit_scene. Times inside a scene use B.lb(i, beat); SFX use project time B.pt(i, beat).
  • Library: text rect circle path_shape line polyline check image group pivot shadow, key fade rise pop show hide snap_in typing counter stepped scramble redact, pill offset tokens_line logo_lockup Thread bubble, sfx music write_doc. Read lib/motion.py docstrings.
  • Fonts: keys head body bodym mono monom (Geist, default), outfit outfitm jbmono, type typeb (Courier Prime). Remap with setup(ROOT, fonts={"head": FONT_FILES["outfit"]}).
  • Rules that bite (references/quirks.md): fade Shapes through a group; keep text lines ≤ 900 px; layers[0] is on top; text y is the baseline; no ✓ glyph (use check()); 1 ms offsets for coincident HOLD keys.
  • Default 1080x1350, light theme, one accent. Dark only if the brief/reference calls for it.
  • Mock content (sample names, counts, code) is fine; real-looking numbers that are not from the user must be listed as illustrative in the final message.
videos/<slug>/.tesseract-work/rebuild.sh                     # ~15 s; prints WARN lines, fix them
bash $POST_TO_VIDEO/scripts/check.sh videos/<slug>             # 20 evenly spaced frames -> checks/fs-a.png, fs-b.png
bash $POST_TO_VIDEO/scripts/check.sh videos/<slug> 1500 2400 …  # or specific ms

Open both filmstrips (Read) and fix: overlaps, text wrapping, empty halves of the frame, things appearing before/after they should, colours on colours (an orange stamp on an orange card disappears), labels colliding. Use "$T" preview --project … --time <s> --output x.png for a full-size frame. Iterate until clean. Usually 1–3 passes.

5. Export and deliver

bash $POST_TO_VIDEO/scripts/export.sh videos/<slug>            # run in background: ~60-80 s; prints streams, LUFS, peak; writes Previews/Filmstrip.png

Then:

  1. Write videos/<slug>/.tesseract-work/NOTES.md: brief, seed patterns → concept, look, music + in-point
    • beat (estimated, not listened), scene beats, real vs. illustrative data, copy changes.
  2. Add a row to references/catalog.md (and music.md "Used by").
  3. Send the MP4 with SendUserFile (display: render).
  4. Final message: the concept in 2–4 lines, scene-by-scene list, audio numbers, and a flag list: copy changes, claim/data gaps, illustrative content, "beat sync not confirmed by ear".

Revisions: copy the current mp4 + build.py into Versions/ before a redesign, then edit build.py and re-run rebuild → check → export.

About Graphed

Graphed is a marketing cloud. It includes the infrastructure for an agent to do marketing: a data warehouse, data pipelines, a Postgres database, cron jobs, runtime servers, a tool catalog, custom domains, and internal app tooling.

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.