What it does
Reads incoming comments and flags likely spam, link-drops, and toxic replies with a reason attached, so a human only has to review the borderline cases instead of reading every comment from scratch.
Install
This skill isn't packaged in an installable registry. It's a plain SKILL.md file you can copy directly. Copy it from the SKILL.md tab and save it at the path for your assistant:
Use it in your AI assistant
Claude Code
.claude/skills/comment-moderator/ (or ~/.claude/skills/comment-moderator/ for all projects)
GitHub Copilot (VS Code)
.github/skills/comment-moderator/
OpenAI Codex
.codex/skills/comment-moderator/
Cursor
.cursor/skills/comment-moderator/ (or ~/.cursor/skills/comment-moderator/ globally)
Compatibility
Any WordPress site with comments enabled. Read access to the comment queue; write access only to set comment status (approve/spam/trash), never to edit comment content.
The complete SKILL.md for this skill:
---
name: comment-moderator
description: Use when a user wants help triaging a comment moderation queue, or asks to find spam, toxic, or off-topic comments among pending comments.
---
# Comment Moderator
## When to use
The comment moderation queue (pending, or all recent) needs triage, separating clear spam and toxicity from legitimate discussion that just needs a human glance.
## Procedure
1. Read pending (and optionally recently-approved) comments with their author name, email domain, and any links included.
2. Classify each into: Clear spam (link farms, unrelated product pitches, gibberish), Likely toxic (personal attacks, slurs, harassment), Borderline (sounds off but could be genuine, sarcasm, strong disagreement, non-native phrasing), Legitimate.
3. For "Clear spam" and "Likely toxic", propose the action (spam / trash) with a one-line reason each; never auto-execute without approval.
4. Leave "Borderline" and "Legitimate" entirely alone, do not auto-approve, since false positives on real commenters are more costly than a slower queue.
5. Summarize counts by category so the user can see queue composition at a glance before opening the moderation screen.
## Guardrails
- Never permanently delete a comment; only ever propose changing its status, and only spam/trash (both reversible), never bulk-execute without a human confirming the batch.
- When genuinely unsure, default to "Borderline", a missed spam comment costs nothing; a wrongly-flagged real commenter costs trust.