gw-media-content-now

Produce one unpublished post piece (texts per channel plus images) in a dated folder, from a proposal id or a free topic. Use when the user picks a proposal from gw-media-content, or asks for a single

SKILL.md

GitHub

gw-media-content-now

If .claude/gw/gw-media-content-now.md exists in the project, read it first: it holds what is specific to the project. Then read .claude/gw/gw-media-content.md if it exists, and its shape in .claude/skills/gw-media-content/nota.md (folder <dir>, languages, channels, extra channels, palette, screenshots, task prefixes, Reddit rules).

You make one piece: <dir>/AAAA-MM-DD/NN-<tema>/, with texts for each channel and the images that go with them. Pieces accumulate in the day's folder. Nothing is ever published: the user posts and records where. Shared material is in .claude/skills/gw-media-content/: catalogo.md, reglas.json, nota.md, plantilla/, scripts/.

Steps

  1. Input. A proposal id (AAAA-MM-DD/P3: read its block in <dir>/propuestas/AAAA-MM-DD.md) or a free topic (pick the row of catalogo.md that fits; with no repo data behind it, say so and stop). Done when you can name the topic, the facts to use and the channels: the proposal's networks (or the request's) among the active ones.
  2. Folder. <dir>/AAAA-MM-DD/ for today; NN is the next free two-digit number there (01 if none); <tema> is a lowercase slug (a-z, 0-9, -). Copy .claude/skills/gw-media-content/plantilla/ into NN-<tema>/ and shape it:
    • delete the channel files not in use; for an extra channel of own format, copy its template (note) as <canal>.md;
    • in ficha.md set Idiomas: (the first language is the main one) and ## Canales, delete the ## Reddit block when reddit is not a channel, and delete the example rows of ## Hechos comprobados and ## Imágenes;
    • replace every <tema>, AAAA-MM-DD and {{...}} placeholder in the folder, or delete its line. Done when the piece folder holds ficha.md plus one file per channel, with no placeholder left.
  3. Facts first: ficha.md. Every claim in the piece goes in the table Hechos comprobados with its source. Numbers come from python3 .claude/skills/gw-media-content/scripts/cifras.py [--desde REF]; the rest from a cited file and line or commit. Only what is merged; if the project has a canon (lore, design docs), claims about it come only from there. Task ids, branches and internal notes go in Uso interno. Done when each fact a channel will use has a source cell.
  4. Media. Every image a channel carries exists in medios/ and is listed in ## Imágenes with alt text in every language; a text-only piece leaves ## Imágenes with no rows.
    • From the template: mv (not cp) the source for the chosen format in medios/fuente/ to its final name, delete the other template sources, and fill every {{...}}. Names per language: the first language medios/fuente/<n>.html; any other language that needs its own image medios/fuente/<n>.<lang>.html (never a suffix for the first language: comprobar.py rejects it). Apply the palette from the note.
    • Screenshots: steps in the note; crop to what matters, save in medios/fuente/ as .png or .jpg; no personal data, no placeholder or broken text on screen.
    • Render: node .claude/skills/gw-media-content/scripts/render.mjs <piece-folder> (it names what is missing and the command to fix it). From medios/fuente/<n>.html it writes medios/<n>.webp, medios/<n>-mini.webp and medios/<n>.png; from <n>.<lang>.html, medios/<n>.<lang>.webp (+ -mini, .png); a carrusel*.html gives a .pdf. Use the .png for LinkedIn (it takes .png, .jpg, .gif, not WebP), the .webp for X, Reddit and Discord, .webp, .png or .jpg for an extra channel, the .pdf carousel for LinkedIn only. Read each render image before using it.
  5. Channels. Write each channel file in every language, from the ficha only, in the shape of its template file and within reglas.json (comprobar.py enforces it). Each post names its image on an Imagen: line; in an own-format channel there is no Imagen: line: the image path (medios/x.webp) appears in the file's text.
    • Reddit (## Reddit in ficha.md): the - **Subreddit:** r/<x>, - **Normas leídas:** AAAA-MM-DD and - **Autopromoción:** lines go before the first ###, then one ### Normas de r/<x> per subreddit holding the rules the user pasted (at least reddit.normas_min_caracteres characters, no # headings inside: paste them as plain lines). Normas leídas: is counted from the piece's date folder and is at most reddit.normas_dias days old. Take the rules from the note or the ficha when they are dated within that limit; otherwise ask the user to paste them (Reddit cannot be read automatically, and rules are never invented), then continue.
    • Extra channels (declared in the note): one <canal>.md as the note says: a section per language within its límite, or the copied own-format template filled in. If the note gives neither, ask.
  6. Check. Build the command from the note as nota.md says (--prefijos-tarea, and --limite <canal>=N or --formato-propio <canal> for each extra channel of the piece) and run python3 .claude/skills/gw-media-content/scripts/comprobar.py [options] <piece-folder>; it prints OK, and each line it prints is a fix to make. Then reread the English: natural, not a literal translation. Done when OK is printed and the reread found nothing to change.
  7. Close. Mark the proposal **Status:** done → <piece folder>. Commit with the project's commit format; the log row names the folder, its channels and that it is ready to publish. Show the user the texts and image paths. After publishing, the user fills ## Publicado in ficha.md.