Clean Spam Replies on X with This Agent Skill

Where this fits

You run a popular X/Twitter account and wake up to a flood of emoji-laden replies from 同城面付, 寻固炮, and other spam accounts. They're not just annoying—they bury genuine engagement and make your posts look like a ghost town. This skill automates the cleanup: it scans the last 7 days of replies, flags likely spam, and lets you hide and mute those accounts via the X API—without touching real comments.

Why agents benefit

• Dry-run first: The skill outputs a JSON list of flagged and borderline replies before any destructive action, so you can audit what the agent is about to do.
• Human-in-the-loop for edge cases: The heuristic misclassifies sarcastic human replies (e.g., "机器人🤖") and spam variants with invisible characters. The skill forces Claude to review borderline items individually, ensuring only true spam is removed.
• Uses only available API endpoints: X removed the block endpoint, so the skill correctly relies on hide reply and mute—both still supported. No wasted calls to dead endpoints.
• ID-based application: After review, you can pass a specific list of reply IDs to apply the cleanup, giving you full control over the final set.
• Works within X's search window: It covers exactly the last 7 days, matching the recent-search API limit, so you don't miss recent spam or waste time on older threads.

Practical scenarios

Cleaning a viral thread overnight

Your post got picked up by an algorithm, and spam accounts swarm in with fake promises. Run the dry-run, review the borderline list, then apply—all in a few minutes, without scrolling through hundreds of replies.

Handling a recurring spammer pattern

You notice the same set of accounts posting identical emoji combos under every post. The skill's heuristics catch these flagged items instantly, and you can mute them all in one --apply run.

Auditing before a public report

If you need to document spam activity for a report or moderation review, the dry-run output gives you a structured JSON list of flagged and borderline items—useful evidence without extra tooling.

Add it to your agent workflow

Install the skill with:

npx skills add jianshuo/claude-skills --skill wjs-cleaning-spam

Then run the dry-run script to see what the agent would clean:

python3 ~/.claude/skills/wjs-cleaning-spam/scripts/clean_spam.py

Review the borderline list (Claude can help you identify spam traits like decorative emojis, 同城 phrases, or invisible Unicode), then apply with --apply or --ids for a curated set.

Compare before adopting

Before adding this skill, compare it against alternatives by checking: quality signals—does the dry-run output match your own judgment on a test thread? maintenance freshness—when was the last commit, and does it still use the current X API? alternatives—are there generic spam-filtering skills that might be more flexible? workflow fit—does the 7-day window and ID-based approach align with how you manage your account?

Why it is worth tracking

This skill is a practical example of a narrow, well-scoped agent capability: it does one thing (clean spam replies) and does it carefully with a human-review step. With 129 stars and an active author, it shows community interest in real-world social media maintenance. Keep an eye on it if you manage high-traffic X accounts or want a template for building similar moderation tools. For a full catalog of such skills, see OpenAgentSkill.