Skip to content

ever-just/agentskills / skills/everjust-website-blog

everjust-website-blog

Verified

EVERJUST Website Blog — Agent Skill

Author and publish blog articles on a live everjust.app tenant's public website: write a post's HTML content, teaser, cover image, tags and SEO/OpenGraph meta, then publish it (now, back-dated, or scheduled) so it renders at /blog on the site — all through the Odoo MCP / ORM (search, get, create, update, call, describe_model; see [[everjust-agent-mcp]] for opening the session against the right tenant DB).

The crucial everjust fact: this is STOCK Odoo 19 website_blog — there is no custom everjust blog addon. The models are vanilla blog.blog / blog.post / blog.tag / blog.tag.category. So read fields from the live model (describe_model / fields_get), not from memory of an older Odoo. What you must know that is everjust/version-specific:

  • A blog POST is a blog.post RECORD, not a website.page. The MCP's WEBSITE TOOLS (website_pages / website_new_page / website_edit_page / website_publish / website_menu) operate on static QWeb pages (ir.ui.view / website.page). A blog.post is rendered by the website_blog controller at /blog/<blog>/<post>, so you author it with create/update on blog.post and publish it with is_published — NOT with website_publish. Using website_publish on a blog article is a category error (recipe 0, Pitfall 1).
  • Publishing is is_published + a post_date that is in the PAST. A post shows on the public /blog only when website_published is True AND post_date <= now. A future post_date is how you SCHEDULE a post (it stays hidden to the public until that time even while published). This build has NO date_publish, NO website_indexed, NO visibility field on blog.post — do not reach for those; the two levers are is_published and post_date.
  • Publishing EMAILS the blog's followers. Setting is_published=True fires the stock website_blog "Published Post" notification (subtype website_blog.mt_blog_blog_published, template website_blog.blog_post_template_new_post) to everyone following the parent blog.blog. This is stock mail.thread chatter over the tenant's outgoing ir.mail_server — NOT the [[everjust-mail-ops]] webmail stack. Know who follows the blog before you publish (recipe 5).
  • Cover image lives in cover_properties (a JSON text blob), not a plain field. The cover is the background-image key inside that JSON, plus color/opacity/height keys.
  • Requires website-designer rights. Creating/publishing posts and (especially) editing the site chrome needs the connected user in website.group_website_designer (a.k.a. can_publish on the post). A plain user can draft but often can't publish. Check your_access / can_publish first.

You reach the ORM through the platform's Odoo MCP tools — see [[everjust-agent-mcp]] for opening the session against the right tenant DB. Every recipe below is expressed through that toolset.

When to use this skill

  • Author a blog post — create a blog.post with title, subtitle, rich HTML content, a teaser, a cover image, tags, and SEO/OpenGraph meta.
  • Publish / unpublish / schedule a post — flip is_published, back-date its post_date so it looks like it went out earlier, or set a future post_date to schedule it.
  • Edit an existing post — rewrite its content, swap its cover, retag it, tune its meta.
  • Manage the blog taxonomy — create/rename blog.tags (and optional blog.tag.category groupings) and attach them to posts so the /blog tag filter works.
  • Set up the blog itself — create the blog.blog container (name, subtitle, cover), and ensure a /blog menu item exists so visitors can reach it.
  • Diagnose — a post that won't appear on /blog (unpublished, future post_date, wrong website_id, wrong blog), or a taxonomy that isn't filtering.

Do NOT use this skill for, and stop if the task is really:

  • Editing the STATIC /blog index chrome, the marketing pages, or any non-blog page. Those are QWeb ir.ui.view / website.page — the everjust marketing pages (website_connectdomain, website_tcsw) are verbatim Tailwind-utility QWeb wrapped in <t t-call="website.layout">. Edit those with the WEBSITE TOOLS (website_edit_page, COW-into-a-site-view) keeping the existing Tailwind idiom — that's the site-chrome path in [[everjust-platform]], NOT this skill. This skill is the content model (blog.post), not the page template.
  • Sending the "new post" announcement as marketing email / a newsletter blast. The publish notification is stock chatter to blog followers; it is not the [[everjust-mail-ops]] webmail product and not a mass-mailing. To email a broad audience about a post, that's a different job (mass_mailing is gated; see [[everjust-platform]]).
  • Event pages / agendas — those are event.event / website_event; see [[everjust-events]]. A blog post is not an event.
  • Comments moderation / the discussion under a post — those are mail.message (message_type='comment') on the post's chatter; managing them is native Odoo chatter, not part of authoring.

Architecture (the model map)

Everything is per-tenant DB and per-website_id (a post can be scoped to one site, or left global). Real fields below are from fields_get on the live connectdomain DB (Odoo 19). Note the build-specific gaps — this build has NO blog.post.date_publish, NO website_indexed, NO visibility/visibility_password on the post; publishing is is_published + post_date only.

ModelRoleKey fields (real, confirmed live)
blog.blogThe blog container — one section of the site holding many posts, with its own landing cover.name* (Blog Name), subtitle, content (Html — the blog's own intro copy), active, website_id (→website; False = show on ALL sites — the default here), cover_properties (JSON cover, same shape as post), blog_post_ids (→blog.post), blog_post_count (computed). Inherits SEO meta (website_meta_*).
blog.postOne article. What you author + publish.name* (Title, required), subtitle (Sub Title — also the default SEO/OG description), blog_id* (→blog.blog, required — which blog it lives in; defaults to the first blog), content (Html — the article body; sanitize=False, translate=html_translate), teaser (computed preview) / teaser_manual (the field you WRITE to override the auto teaser), tag_ids (→blog.tag m2m), author_id (→res.partner, default = you), author_name (stored, from author), cover_properties (JSON text — the cover image + styling), is_published (bool, stored — the publish flag you set), website_published (bool, computed/non-stored mirror — read it, don't write it), post_date (Datetime — the public "publishing date"; a FUTURE value hides the post = scheduling), published_date (Datetime — when it was actually first published), can_publish (bool computed — may YOU publish?), website_id (related from blog_id.website_id, read-only — set it via the blog, not the post), website_url (computed slug URL /blog/<blog>/<post>), visits (view count, readonly), active, website_meta_title / website_meta_description / website_meta_keywords / website_meta_og_img / seo_name (SEO/OpenGraph overrides).
blog.tagA taxonomy tag for filtering /blog.name* (required, UNIQUE across the tenant), category_id (→blog.tag.category, optional grouping), color (int, kanban color), post_ids (→blog.post m2m). Inherits SEO meta.
blog.tag.categoryOptional grouping of tags (e.g. "Topic", "Product").name* (required, UNIQUE), tag_ids (→blog.tag).

The publish gate — read the codes, not the labels

A post is publicly visible on /blog only when BOTH hold:

  1. website_published is True — which you set by writing is_published=True (is_published is the stored source; website_published is the non-stored mirror the front-end reads). AND
  2. post_date <= now — the public _search_get_detail domain filters ("post_date","<=",now) for non-designers. A future post_date on a published post keeps it hidden from the public until that datetime — this is exactly how you schedule a post. (A logged-in website-designer can preview unpublished/future posts; the public cannot.)

So the two levers are is_published (bool) and post_date (Datetime). There is no separate "schedule" field in this build — set post_date to the future to schedule, to the past to back-date. published_date records when it first went live and is generally managed for you (the model stamps it on first publish); don't fight it — control visibility via is_published + post_date.

What is_published=True triggers (it emails followers)

blog.post.write/create call _check_for_publication: on transition to published it posts the website_blog.blog_post_template_new_post message with subtype website_blog.mt_blog_blog_published on the parent blog.blog — which notifies the blog's followers (blog.message_partner_ids) by email through the tenant's stock outgoing ir.mail_server. This is NOT silent and NOT the [[everjust-mail-ops]] webmail stack. Before publishing, know who follows the blog (recipe 5) so you don't spray an unexpected announcement — or unpublish-then-set the followers first. Repeated re-publish (already published → write is_published=True again) does not re-fire the "new post" message for the same transition.

The cover image is JSON in cover_properties

cover_properties is a text field holding a JSON object, e.g. {"background-image": "url('/web/image/blog.post/12/cover_image')", "background_color_class": "o_cc3", "opacity": "0.2", "resize_class": "o_half_screen_height"}. The cover picture is the background-image key (value none = no cover, or url('…')). To set a cover you (a) upload the image as an ir.attachment / image and reference its URL, then (b) write the whole JSON blob back into cover_properties. resize_class controls banner height (o_half_screen_height / o_full_screen_height / cover_mid_height / o_record_has_cover), background_color_class the overlay color, opacity the dim. Don't write a bare URL into cover_properties — it must be valid JSON with the background-image key.

SEO / OpenGraph

subtitle is the DEFAULT OG/Twitter description and meta description if you set nothing else. To override: website_meta_title (page <title> / og:title), website_meta_description (og:description), website_meta_keywords, website_meta_og_img (explicit OG image URL — else the cover's background-image is used), seo_name (the URL slug override). OG article:published_time = post_date, article:tag = the tag names — so tag and date drive OG automatically; you mostly just set title/description/og-image.

Recipes

Route each call through the Odoo MCP against the tenant DB (open the session per [[everjust-agent-mcp]]; pick the DB, e.g. connectdomain). Recipes use the MCP toolset (search, get, create, update, call, describe_model); ORM env[...] forms are shown where clearer. On connectdomain at last check: one blog "Connect Domain Blog" (id 2, website_id=False ⇒ all sites), 4 posts (all is_published=False), a /blog menu already present, and 0 tags.

0. First: confirm this is the blog CONTENT path, not the page path — and that you can publish

The single most important orientation. A blog article is a blog.post record; the WEBSITE TOOLS are for static pages. Do NOT reach for website_new_page/website_publish to make an article.

describe_model('blog.post')     # confirm your_access.{create,write}; note fields
# The blog(s) that exist on this tenant (which website each is bound to):
search('blog.blog', [], ['name','subtitle','website_id','blog_post_count'])
# Can YOU actually publish (vs only draft)?  can_publish needs website-designer rights:
get('blog.post', [<existing_post_id>], ['can_publish','is_published','website_published'])

If can_publish is False / your_access.write is False, you are not in website.group_website_designer — you can draft but not publish; do not retry with confirm:true (it's an ACL limit, not a confirm gate — see [[everjust-agent-mcp]]). To reach the static /blog index chrome or a marketing page instead, that's website_pages / website_edit_page (the QWeb path in [[everjust-platform]]), NOT this recipe.

1. Author a draft post (content + teaser + subtitle), unpublished

Create it unpublished first (is_published=False) so you don't email followers before it's ready. content is HTML (sanitize=False) — write real article markup.

pid = create('blog.post', {
    'blog_id': 2,                                  # -> the target blog.blog (resolve via recipe 0)
    'name': 'Bring Your Own Domain, Automatically',   # Title (required)
    'subtitle': 'DNS auto-config + automatic SSL, without the support tickets.',  # also default OG desc
    'content': '<p>Custom domains used to mean a support queue…</p>'
               '<h2>How it works</h2><p>…</p>'
               '<figure><img src="/web/image/…" alt="…"/></figure>',
    'teaser_manual': 'A 1–2 sentence hook shown on the /blog listing card.',   # override auto teaser
    'is_published': False,                         # draft — no follower email yet
})
get('blog.post', [pid], ['name','website_url','is_published','post_date','teaser'])

Leave teaser_manual empty to auto-derive the teaser from the first ~200 chars of content. The website_url returned is the slug path (/blog/<blog-slug>/<post-slug>-<id>). author_id defaults to your partner; set it explicitly if the byline should differ. Don't set website_id on the post — it's related from blog_id.website_id; choose the site by choosing the blog.

2. Give it a cover image (the cover_properties JSON)

The cover is the background-image key inside cover_properties. Upload the image, then write the whole JSON back.

# a) stage the image (an attachment you can serve):
att = create('ir.attachment', {
    'name': 'byod-cover.png',
    'datas': '<base64>',                 # the PNG/JPG bytes, base64
    'res_model': 'blog.post', 'res_id': pid,
    'mimetype': 'image/png', 'public': True,
})
url = '/web/image/%s' % att            # servable URL for that attachment

# b) write the cover JSON (background-image + banner styling):
import json
update('blog.post', [pid], {'cover_properties': json.dumps({
    'background-image': "url('%s')" % url,
    'background_color_class': 'o_cc3',
    'opacity': '0.2',
    'resize_class': 'o_half_screen_height',   # banner height
})})

background-image: "none" = no cover. Keep it valid JSON — a bare URL string breaks the template. The cover doubles as the OG image unless you set website_meta_og_img explicitly (recipe 4).

3. Tag the post (create the taxonomy if it's missing)

blog.tag.name is unique — reuse an existing tag id, don't create a duplicate name. On connectdomain there are currently 0 tags, so you'll create them.

# find-or-create tags (name is unique):
def tag(name):
    hit = search('blog.tag', [['name','=',name]], ['id'], 1)
    return hit[0]['id'] if hit else create('blog.tag', {'name': name})
tids = [tag('Product'), tag('DNS'), tag('Announcements')]
# optional: group tags under a category
cat = create('blog.tag.category', {'name': 'Topic'})     # name unique
update('blog.tag', tids, {'category_id': cat})
# attach to the post (m2m command tuple: 6,0 = replace set):
update('blog.post', [pid], {'tag_ids': [[6, 0, tids]]})

Tags power the /blog tag filter and feed OG article:tag automatically. Use [[4, id]] to add a tag without replacing the set, [[3, id]] to remove one.

4. Set SEO / OpenGraph meta

If you set nothing, subtitle becomes the meta/OG description and the cover becomes the OG image — often good enough. Override when you want a distinct search title/snippet:

update('blog.post', [pid], {
    'website_meta_title': 'Bring Your Own Domain for SaaS | Connect Domain',
    'website_meta_description': 'Let users connect a custom domain with automatic DNS + SSL. '
                               'Here is how Connect Domain does it in one click.',
    'website_meta_keywords': 'custom domains, BYOD, SaaS, DNS, SSL',
    'website_meta_og_img': url,          # explicit OG image (else the cover is used)
    'seo_name': 'bring-your-own-domain',  # URL slug override
})

og:type='article', article:published_time (=post_date), article:modified_time, and article:tag (=tag names) are filled by the model — you don't set them. Keep the meta title ≲60 chars and description ≲155.

5. Publish — now, back-dated, or scheduled — after checking who gets emailed

Publishing emails the blog's followers (see Architecture). Look before you publish.

# Who follows the blog (they get the "Published Post" email)?
get('blog.blog', [2], ['message_partner_ids'])       # -> partner ids following the blog
search('mail.followers', [['res_model','=','blog.blog'],['res_id','=',2]], ['partner_id'])

# a) PUBLISH NOW (visible immediately; fires the follower email):
update('blog.post', [pid], {'is_published': True, 'post_date': '2026-07-05 14:00:00'})  # <= now

# b) BACK-DATE (looks like it went out earlier; still visible now since past<=now):
update('blog.post', [pid], {'is_published': True, 'post_date': '2026-06-20 09:00:00'})

# c) SCHEDULE (published flag set, but hidden from public until a FUTURE post_date):
update('blog.post', [pid], {'is_published': True, 'post_date': '2026-08-01 08:00:00'})
#    -> public cannot see it until 2026-08-01 08:00 UTC; designers can preview it.

# UNPUBLISH (pull it from the site; does not delete):
update('blog.post', [pid], {'is_published': False})

post_date is a Datetime, UTC-naive (like the rest of Odoo) — a future value is the schedule. is_published is what you write; website_published is the computed mirror you read to confirm. If you must publish quietly, note there is no per-write "skip notification" flag on this path — remove unwanted followers from the blog.blog first, or publish while the post is already the current state. Prefer this over the WEBSITE TOOLS: website_publish publishes a website.page, not a blog.post — it will not make your article live (Pitfall 1).

6. Set up the blog container and its /blog menu (only if missing)

Usually the blog + /blog menu already exist (they do on connectdomain). Create only if a tenant has none.

# a blog exists?
search('blog.blog', [], ['name','website_id'])
# create one (website_id=False shows it on all sites; set an id to scope to one site):
bid = create('blog.blog', {
    'name': 'Connect Domain Blog',
    'subtitle': 'Custom domains, done automatically.',
    'content': '<p>Notes from the team building open-source custom-domain onboarding.</p>',
    # 'website_id': 1,          # optional: scope to a single site
})
# ensure a nav menu points at /blog (the WEBSITE TOOL is right here — it's site chrome, not content):
website_menu(list=True)                        # is there already a /blog item?
website_menu(name='Blog', url='/blog', sequence=30)   # add one if not

The /blog listing page and menu are site CHROME — that's the one place the WEBSITE TOOLS (website_menu, and website_edit_page if you want to restyle the index) legitimately apply. The individual article pages are rendered from blog.post records by the website_blog controller — you never create a website.page per article.

Pitfalls (everjust-specific)

  1. A blog POST is a blog.post record, NOT a website.page — publish with is_published, not website_publish. The MCP WEBSITE TOOLS (website_pages/website_new_page/website_edit_page/ website_publish/website_menu) act on static QWeb pages (ir.ui.view/website.page). A blog article is content rendered by the website_blog controller at /blog/<blog>/<post>. Calling website_publish on an article does nothing to make it live. Author with create/update on blog.post; publish with is_published=True + a past post_date. The WEBSITE TOOLS are only for the /blog index chrome, the menu, and marketing pages (site-chrome path in [[everjust-platform]]).

  2. Publishing (is_published=True) EMAILS the blog's followers. It fires the stock website_blog "Published Post" notification (subtype mt_blog_blog_published) on the parent blog.blog to its followers, over the tenant's outgoing ir.mail_server. It is not silent and it is not the [[everjust-mail-ops]] webmail product. Check blog.message_partner_ids / mail.followers before publishing; author as a draft (is_published=False) first and only flip it when the piece is final.

  3. Publish gate = is_published AND post_date <= now. A future post_date HIDES the post. This build has no date_publish/website_indexed/visibility field on blog.post. Scheduling is a future post_date on a published post (public sees nothing until then; designers can preview). Back-dating is a past post_date. Don't hunt for a "schedule" or "sitemap" field — set post_date.

  4. is_published is the stored flag you WRITE; website_published is the computed mirror you READ. Write is_published; read website_published/can_publish to confirm state and rights. Don't try to write website_published (non-stored) or website_id (related, read-only — choose the site by choosing the blog_id).

  5. The cover image is JSON in cover_properties, not a plain image field. The picture is the background-image key inside that JSON blob (url('…') or none). Upload the image, then write the whole valid-JSON object back (with background_color_class/opacity/resize_class). A bare URL string in cover_properties breaks rendering.

  6. blog.tag.name and blog.tag.category.name are UNIQUE — find-or-create, don't duplicate. Search for the tag by name and reuse its id; a duplicate name raises the uniqueness constraint. Use m2m command tuples on tag_ids ([[6,0,ids]] replace, [[4,id]] add, [[3,id]] remove) — never a bare id list.

  7. content is trusted HTML (sanitize=False) — you own what you inject. The article body is not sanitized, so malformed or unsafe markup goes straight in. Write clean, well-formed HTML; don't paste untrusted markup. teaser_manual overrides the auto teaser; leave it blank to auto-derive from content.

  8. can_publish / website-designer rights gate publishing. Creating a draft may be allowed while publishing is not — publishing (and any website_edit_page on chrome) needs website.group_website_designer. If can_publish/your_access.write is False, that's an ACL limit; don't retry with confirm:true. Get a key from a user with designer rights (see [[everjust-agent-mcp]]).

  9. website_id=False on the blog means "all sites"; a set id scopes it. On multi-site tenants, confirm which site a blog/post belongs to before creating — a post inherits website_id from its blog_id. On connectdomain the one blog is global (website_id=False), so its posts show on the single site. Everything is per-tenant DB and per-website_id — verify you're on the right DB (see [[everjust-platform]]).

  10. Don't confuse the article's chatter comments with authoring. Reader comments are mail.message (message_type='comment') on the post's chatter; moderating them is native Odoo chatter, separate from writing/publishing the article. And the publish notification going to followers is chatter too — it does not go through mass_mailing or [[everjust-mail-ops]].

See also

  • [[everjust-platform]] — tenancy model, /odoo debrand, the everjust.public_website gate, per-tenant website/theme installs, and the QWeb site-chrome / marketing-page path (WEBSITE TOOLS, Tailwind-utility QWeb under <t t-call="website.layout">, COW-into-a-site-view). Read it for editing the /blog index or a marketing page — this skill is the blog CONTENT model, not the page template.
  • [[everjust-agent-mcp]] — how to connect to the tenant MCP and the generic search/get/create/update/call/describe_model toolset every recipe above uses, plus the WEBSITE TOOLS' signatures and the confirm/ACL gates.
  • [[everjust-events]] — the sibling public-website content type (event.event/website_event). Events and blog posts both publish to the site but are different models with different publish gates (is_published+post_date here vs is_published+website_menu there) — don't cross them.
  • [[everjust-mail-ops]] — the SEPARATE everjust webmail product (everjust.mail.*). The blog's publish-notification email is stock mail.thread chatter to blog followers, NOT this stack — use that skill only if you're operating the webmail product itself, never to send the post announcement.
slack-gif-creatoranthropics/skillsKnowledge and utilities for creating animated GIFs optimized for Slack. Provides constraints, validation tools, and animation concepts. Use when users request animated GIFs for Slack like "make me a GIF of X doing Y for Slack."Stars: 198kUpdated Claude CodeCodexVerifiedInstall
ponytailDietrichGebert/ponytailLazy senior dev mode for any coding task: the smallest change that fully solves it, and a reply a busy human understands in one read.Stars: 194kUpdated Claude CodeCodexVerifiedInstall
ponytail-debtDietrichGebert/ponytailHarvest every shortcut: comment into one debt ledger, so deferrals get tracked instead of forgotten. One-shot report.Stars: 194kUpdated Claude CodeCodexVerifiedInstall
ponytail-debtDietrichGebert/ponytailList every `shortcut:` comment (and older `ponytail:` ones) as a debt ledger. One-shot report, changes nothing. Use for "ponytail debt", "what did ponytail defer", "list the shortcuts", /ponytail-debt.Stars: 194kUpdated Claude CodeCodexVerifiedInstall