Repository navigation
[pull] main from forem:main - #404
Merged
Merged
Conversation
) * Use Jev as a cheap escalation check for spam without links Article and comment spam checks only reached Gemini (ArticleCheck/CommentCheck) when the content had an <a> tag, so spam with contact details written as text (Telegram, WhatsApp, phone) never got an AI check. Content without links now goes to a Jev check (TypeSafe's classifier) first, and only escalates to the existing Gemini check when Jev flags it. Jev never flags anything on its own. Without TYPESAFE_API_KEY nothing changes. Claude-Session: https://claude.ai/code/session_01FLx46puMiJkQWtNe7kTRJU * Move the spam escalation check onto Ai::TypeSafe and Ai::FunctionConfig Replaces the PR's standalone Ai::Jev client with the shared Ai::TypeSafe::Client, Questions and Result from #23879, so the escalation check gets the same retries, AiAudit logging and model pinning (TYPESAFE_API_MODEL) as every other Jev function. Escalation is now the :spam_escalation AI function, selectable under Config > AI Models. It is Jev-only: its choices are Off (the default) and Jev, so setting TYPESAFE_API_KEY for other functions no longer turns it on implicitly. The spam handler gates on Ai::FunctionConfig.available? for the article/comment spam checks, so escalated content goes to whichever model those checks are set to, Gemini or Jev. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Bound Jev latency in high-priority jobs and make the escalation threshold admin-tunable The spam and moderation jobs (Articles/Comments::HandleSpamWorker, DetectCodeBlockLanguagesWorker) run on high_priority and can make several Jev calls each. With a 20s timeout and 3 retries, a TypeSafe outage could hold a worker for ~87s per call. - Ai::TypeSafe::Client takes timeout: and max_retries:, defaults to the SDKs' 10s timeout, and exposes FAIL_FAST (5s, no retries). Every Jev caller on a high-priority path uses it: ArticleCheck, CommentCheck, SpamEscalationCheck, ContentModerationLabeler, SubforemFinder, ArticleEnhancer and DetectCodeBlockLanguages. Background callers keep retries. - A circuit breaker shared through Rails.cache opens after 5 outage failures (timeouts, connection errors, 429, 5xx) in a minute and skips all Jev calls for 2 minutes. - DetectCodeBlockLanguages aborts when the circuit is open instead of writing "plaintext" for every remaining block. - The spam escalation threshold is a global Settings::AiFunctions value (default 0.3, validated to (0, 1]) editable under Config > AI Models > Tuning. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> --------- Co-authored-by: Ben Halpern <bendhalpern@gmail.com> Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
…by it (#23900) * Run the spam domain check in a worker so banish can't be rolled back by it Granting the spam or suspended role ran Spam::DomainDetector inside the rolify after_add callback. Its `email LIKE '%@Domain'` scan can hit the statement timeout on uncommon domains; the QueryCanceled then rolled back the role insert and aborted Moderator::BanishUser right after it created the BanishedUser row. Honeybadger ignores QueryCanceled, so it failed silently and re-banishing hit the same wall. Claude-Session: https://claude.ai/code/session_018TsZkCc8o2SwEbux3MQgqz * Skip domain detection for already-blocked domains and cover the worker BlockDomainAndSuspendUsersWorker blocks the domain before suspending each user, so each resulting detector job now short-circuits on the indexed blocked_email_domains lookup instead of re-scanning users by email. Claude-Session: https://claude.ai/code/session_018TsZkCc8o2SwEbux3MQgqz
ForemStatsClient calls are forwarded to Datadog unchanged and also written to Better Stack as one structured event each, so monitors built on the app's own metrics can be rebuilt before Datadog is turned off. Inert unless BETTERSTACK_METRICS_SOURCE_TOKEN is set in production. Claude-Session: https://claude.ai/code/session_01JGecM5hifj2QYgVRa7qDm7
…led spam reactions (#23883) * Auto-mark repeat spam authors, fix spam check commit race, log failed spam reactions - Low-trust authors (<4 badges) with 3+ mascot vomits on clear_and_obvious_* posts in the last month get the spam role, without waiting for moderators to confirm each vomit. - Enqueue Articles::HandleSpamWorker after the transaction commits so the worker can find newly created articles and sees the committed published state. - Log when the mascot's spam reaction fails validation instead of dropping it silently. Claude-Session: https://claude.ai/code/session_01FLx46puMiJkQWtNe7kTRJU Entire-Checkpoint: 01M3D7GD03KRTQBATV6Z11S3TE * Count only active auto-flags for repeat offenders, keep the count database-side - Ignore mascot vomits that moderators have invalidated or archived (valid_or_confirmed). - Pass the flagged articles as a subquery instead of loading their IDs into Ruby. - Specs for invalidated flags and for the failed-reaction warning and its quiet case. Claude-Session: https://claude.ai/code/session_01FLx46puMiJkQWtNe7kTRJU * Note why repeat auto-flagged authors are marked as spam, expand spam handler tests Leave an automatic_spam note from the mascot (with the flag count) when a repeat auto-flagged author gets the spam role, skipping it when the author is already spam so queued jobs don't duplicate notes. Add specs for the note, each exclusion rule of the repeat-offender count, and the after-commit enqueue. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> --------- Co-authored-by: Ben Halpern <bendhalpern@gmail.com> Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
This branch had an error being deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
See Commits and Changes for more details.
Created by
pull[bot] (v2.0.0-alpha.4)
Can you help keep this open source service alive? 💖 Please sponsor : )