SOCIAL MEDIA
Toxicity gate before auto-replying to social mentions
When a draft reply to an inbound social mention is generated, it is scored for toxicity by a HuggingFace classifier.
How it runs
The automated pipeline, trigger to output.
- TriggerNew brand mention receivedHTTP webhook
- ActionDraft reply with OpenAIOpenAI
- ActionScore draft for toxicity (HuggingFace)Hugging Face
- LogicToxicity score below ceiling?
- ActionPublish clean replySocial publishing
- OutputRoute flagged draft to Slack for reviewSlack
What it does
This workflow puts a safety checkpoint between your AI reply-drafter and the public internet. Every draft reply to an incoming mention is run through a HuggingFace toxicity classifier. If the draft scores below your threshold it publishes automatically; if it trips the threshold it is held and pushed to a Slack channel for a human to approve, edit, or kill.
When to use it
Use it when you let an assistant respond to mentions at volume but a single off-tone reply would damage the brand. It gives you autopilot speed on the safe majority while guaranteeing a human sees anything risky before it goes live.
How it works
- 1A new brand mention arrives via webhook from your social listening source.
- 2OpenAI drafts a contextual reply to the mention.
- 3HuggingFace scores the draft text for toxicity and returns a probability.
- 4A logic step compares the score to your configured ceiling.
- 5If clean, the reply is published to the originating platform.
- 6If flagged, the draft, score, and original mention are posted to Slack for human review instead of publishing.
Set it up
What you configure once, before turning it on.
- 1Connect HTTP webhookTrigger any URL on agent actions.
- 2Connect OpenAIModels, embeddings, files.
- 3Connect Hugging FaceModels, datasets, spaces — the open-source hub.
- 4Connect Social publishingCross-post to X, LinkedIn, Instagram, TikTok, and 4 more in one call.
- 5Connect SlackChannels, DMs, threads, mentions.
- 6Set each agent's modelWe leave models unset so you pick the tier — fast + cheap, or top-quality.
- 7Tune it to your dataEdit the prompts, filters, and field mappings so it matches how your team works.
- 8Test, then turn it onRun once against a sample, confirm the output, then enable the trigger.
More Social Media workflows
A/B carousel cover variants from one brief
Takes a single approved brief and uses an LLM to write three distinct cover-slide headlines, renders a carousel cover for each, and logs the variants to Airtable for an A/B test.
Localized carousel variants per target market
From one approved master brief, translates the copy into each target locale, renders a localized carousel per market, and publishes each to its region's platform account.
Multi-Platform Comment Escalation Router
Ingests inbound comments and replies from your connected social platforms, classifies each for legal or PR risk.
Post-event Discord attendance recap and host digest
After a synced calendar event ends, reconciles who RSVP'd against who actually showed, posts a thank-you recap in Discord, and emails the host a digest of attendance and no-shows.
Agent-Driven Comment Policy Review Queue
An agent reviews ambiguous comments against your brand's response policy, decides whether to auto-clear, draft a reply for approval, or escalate, and routes its decision…
Carousel render with brand-safety pre-check
Before rendering, an LLM checks the brief copy against brand and policy rules; clean briefs render and archive to S3, flagged briefs route to a review queue instead.
Run it inside a business
This workflow drops into a full company template. Import the org, and this is one of the playbooks its agents run.
Run this workflow in your colony.
14-day trial. No DevOps. No Sales call. Provisioned in under a minute.