Weekly Summary

This week on meta.discourse.org

(Covering 2026-07-26 → 2026-08-02)

Key Stats

  • New posts: 701
  • New topics: 72

Top contributors (posts / likes received):

User Posts Likes
Lilly 40 72
Moin 40 70
awesomerobot 23 53
Damian_Boon 36 52
chapoi 21 47
alltiagocom 40 45
NateDhaliwal 22 42
zogstrip 13 42
merefield 27 41
Canapin 16 38

What drove the conversations this week (high-level highlights): a new ESR release and follow-on upgrade questions (read more), more iteration on nested replies (including a new “Hot” sorting mode) (read more), and a cluster of UX/admin dashboard discussions about “where did the version info go?” (read more).


Interesting Topics

#enterprise

  • How close should customers be to your teams? HAWK kicked off a thoughtful thread on setting healthy boundaries between customers and product teams—covering “hidden costs” of engagement and how to build trust without creating an always-on expectation (read more).

#announcements

  • July 2026 monthly release + ESR shift: The 2026.7 release landed and became the new ESR, prompting practical guidance on pinning to a specific branch (e.g. release/2026.1) to avoid surprise jumps (read more).

  • Intermediate security releases (July 31): david posted the “intermediate releases” roundup and clarified that some “unexpected jumps” were actually due to the new ESR being released earlier in the week (read more).

  • Composer controls cleanup: The composer preview toggle moved to sit alongside the other window controls (fullscreen/minimize), and the minimized draft bar was simplified for clarity (read more).

  • Nested replies: “Hot” sorting: A new sort option designed to surface currently-engaging subthreads in nested replies—bridging the gap between “Top” and “New” (read more).

  • Pending user approval reminders: new default: A new upcoming change drops the default delay from 8 hours to 30 minutes on sites with approval enabled, improving staff responsiveness for queued users (read more).

#extras ai

  • Improving AI report accuracy via validation: sam introduced a more robust pipeline for AI periodic reports by adding workflow-based validation steps—aimed at improving summary quality without changing LLMs (read more).

  • “Agent skills” for Discourse AI: A proposal to store skill definitions in topics (one topic per skill), enabling more maintainable and permissioned agent behavior through progressive disclosure (read more).

  • Workflows + Gemini schema pain: A thread on Gemini choking on schema constructs (notably additionalProperties), with sam noting a spike toward the Interactions API for a more stable bridge (read more).

Community Building

  • Moderating “AI-ish” content (without constant conflict): A practical report on using an AI classifier to flag overly formal / thesis-like AI-generated topics (especially from new users), including real-world cost estimates and moderation workflow details (read more).

  • Office Hours replay: embedded Discourse in full app mode: The team shared a replay about bringing Discourse conversations directly into external websites via full-page embedded experiences (read more).

  • Survey: needs of community administrators: A bachelor’s thesis survey asked admins/mods what analytics actually matter (newcomers, departures, workload, returning users, etc.) (read more).

#self-hosting

  • Cheaper alternatives to Hetzner? A wide-ranging thread on pricing changes, availability constraints, and real-world sizing advice (plus a script idea to alert when specific Hetzner instances become available) (read more).

  • Maintenance page workaround (Cloudflare Workers vs dual-container): Discussion weighed “nice touch” maintenance UX against complexity; dual-container rebuild strategy was repeatedly recommended to shrink downtime windows (read more).

  • “Updated to ESR but 578 new changes available” bug: david explained a DiscourseHub version-check issue (counting commits between refs) and posted the fix + a manual job trigger for impatient admins (read more).

ux

  • Where did the version information go (new dashboard concerns): Admins debated the redesigned dashboard focus (community manager vs sysadmin needs), with discussion on roles/hats and potential future customization (read more).

  • Dark theme readability: reduce pure-white text: A design request to avoid harsh high-contrast white-on-black typography—especially relevant to the newer palettes in Horizon-style themes (read more).

  • Staff-only mode should not show login/signup CTAs: A UX improvement to avoid frustrating anonymous users with actions they can’t complete—fixed via a core PR (read more).

  • “Start Updating” button doesn’t look clickable: A small-but-painful affordance issue in the admin update UI, with a note that it changes under certain upcoming theme changes (read more).

bug

  • Turkish translation outputs archaic/Ottoman language: The “Translate to Turkish” feature was over-transforming; the prompt was tightened to “stay in its lane” and produce faithful translations (read more).

  • Chat search results can’t scroll on mobile: Global chat search results were unscrollable via touch; fix shipped quickly in core (read more).

  • Homepage shows older topics (bfcache headers): A long-running “latest isn’t latest” issue was traced to bfcache-compatible headers and addressed via a PR (read more).

  • Reply-by-email from Outlook truncates after numbered lists: A reproducible formatting bug (with fixture email) was fixed via PR (read more).

development (and closely related implementation threads)

  • Trust Level 3 transparency: progress UI + core logic clarification: A plugin to expose TL3 progress to TL2 users sparked a broader conversation about what to reveal vs what can be gamed (read more), while a separate deep dive clarified how the % posts read requirement and the cap work in core (read more).

  • Logo behavior: small logo breakpoint override: A “why does the primary logo stay too long?” question ended up traced to a theme component overriding behavior; a new setting was added to restore default behavior (read more).

  • Correct CSS for light/dark: use light-dark() A concise styling tip: the native CSS light-dark() function can be simpler than container queries when you truly need explicit dual colors (read more).

#plugin / #theme-component

  • Resenha voice rooms: rapid UX iteration + new features: Multiple improvements landed (private lock icon, participant list discoverability, deep links to rooms), plus highlights like coturn auth, recording, flags, and tile right-click menus (read more).

  • Theme component: Mobile Composer Preview restored: The preview toggle was brought back for mobile (and debugging uncovered some Theme Creator preview-link flakiness, with better testing workflow tips) (read more).

  • Plugin: Event Duplicator for calendar topics: A new plugin to duplicate event topics into a new time period gained “shift (months)” support quickly (useful for monthly/quarterly event patterns) (read more).


Activity by the @team Group

sam

  • Pushed forward Discourse AI reporting quality by adding workflow-driven validation steps to the AI report pipeline (read more).
  • Proposed a structured approach to “agent skills” stored as topics—aimed at reducing unwieldy system prompts and making agent capabilities easier to maintain collaboratively (read more).
  • Responded to Workflows + Gemini provider limitations and noted an upcoming push toward Interactions API for better stability (read more).
  • Shipped (or pointed to) fixes around media review behavior and edge cases in upload review logic (read more).

david

  • Clarified ESR upgrade mechanics (pinning to release/2026.1 vs tracking esr) in the monthly release thread (read more).
  • Posted the intermediate security releases list and explained why some “stable” installs jumped due to ESR timing rather than the intermediate release itself (read more).
  • Diagnosed and resolved the “578 new changes available” display issue as a DiscourseHub version-check API bug (read more).
  • Traced an avatar/image optimization failure back to an archived plugin and recommended removing it to restore normal uploads (read more).
  • Helped narrow down remaining iOS upload problems (beyond the older-iOS fix) by requesting better UA/log evidence (read more).

Falco

  • Fixed an AI translation prompt that was over-stylizing output, causing archaic Turkish; the revised prompt keeps translations faithful to tone and meaning (read more).
  • Delivered multiple Resenha improvements (UX fixes for private rooms and better participant management discoverability), plus deep-linking to rooms (read more).
  • Investigated iOS upload issues, suspecting instance-specific settings when “sandbox works but my site doesn’t” (read more).
  • Reduced noise from AI “structured output JSON warning” logs by moving that message to INFO level (read more).

zogstrip

  • Improved UX for staff-only sites by removing misleading login/signup CTAs for anonymous users (read more).
  • Introduced the “pending users reminder default” upcoming change, reducing the delay to 30 minutes for sites requiring approval (read more).
  • Landed multiple fast-turnaround fixes across moderation/review and notifications:
    • “Suspend” option in review flows / consistency improvements (read more)
    • Duplicate poll title in email notifications (read more)
    • Flag “reason” formatting regression fix (read more)
    • Outlook reply-by-email truncation fix (read more)
  • Traced “opening forum shows older topics” to bfcache-compatible headers and posted a PR to address it (read more).
  • Implemented the requested boolean input behavior for manually mapped Workflows fields and kept the topic updated with PR links and status (read more).

chapoi

  • Announced a composer control tidy-up (preview toggle moved, minimize icon standardized, draft bar simplified) (read more).
  • Helped users navigate support and UX behavior around unread/unseen semantics (post-level vs topic-level) (read more).
  • Guided translation contributors on the correct path to become proofreaders (and linked the canonical instructions) (read more).
  • Confirmed Horizon’s intentional declutter decision re: hiding the “unread/new topics” line under suggested topics (read more).

awesomerobot

  • Shipped a fix for mobile chat global-search scrolling and linked the core PR for inclusion on next update (read more).
  • Identified theme-component interference in header logo swapping behavior and added a new setting to restore default behavior in the full-width component (read more).
  • Advised on safer customization patterns (CSS variables over brittle selectors, stable plugin outlets over overrides) and what tends to break across updates (read more).
  • Provided a clean solution for light/dark styling using the CSS light-dark() function (read more).
  • Helped debug URL anchor stripping issues in the rich editor by suggesting HTML wrappers that survive composer transformations (read more).
  • Flagged a Meta theme translation issue (empty banner text) and looped in nat for investigation (read more).

HAWK

  • Started a substantial discussion on staff participation frameworks and reducing fear-of-engagement across teams (especially when not everyone is an engineer/designer) (read more).
  • Weighed in on AI-content moderation norms and pushed back (lightly, but clearly) on the idea that humans should “downgrade writing style” to prove they’re not bots (read more).
  • Pointed users toward site text customization as the first stop for notification wording changes (read more).

markvanlan

  • Introduced “Hot sorting” for nested replies—designed to surface active subthreads without freezing early-liked replies at the top forever (read more).
  • Confirmed expected behavior for “new nested reply appears at top until refresh” and explained the UX rationale (visibility vs strict ordering) (read more).
  • Continued nested replies stabilization work with fixes for scroll jumping and post leakage across topics (read more).

mcwumbly

  • Explained the TL3 “posts read” math and the cap in a way that finally made the logic click for folks reading core code (read more).
  • Added nuance to the “where did version info go?” debate by framing admin needs as multiple “hats” (sysadmin vs site admin vs community manager) and pointing toward more customizable /admin futures (read more).
  • Turned a “Me too” reporting question into a working Data Explorer query for admins to track shared-issue clicks by topic + date range (read more).
  • Engaged in early thinking around nested subcategory presentation in dropdowns, including tradeoffs and potential UI affordances (read more).

nat

  • Fixed Meta’s “empty banner box” issue by removing accidental manual empty overrides in a theme component’s translations (read more).
  • Merged a date picker improvement for AI Translate settings to make backfill dates clearer and easier to configure (read more).
  • Advanced multi-language preferences for localized content display, sharing screenshots and inviting Meta testing/feedback (read more).

Thanks for reading, and I’ll see you again next week! :slight_smile: