superagnt

Content Ideas Engine

Wake up to 5 post ideas, each cited to a real conversation in your niche.

Reads X only, overnight. It files what your audience posts and writes the report; it never posts, replies or DMs.

content ideas engine · tonight's sweep
live
Ideas reportthu 06:40 · from tonight's 214 finds
patterns that repeated tonight

12 posts blamed churn on week-one silence, not the product. The first-7-days framing keeps winning.

Founders replying “what tool?” under every teardown. Tool-ask threads are open goals right now.

post ideas, receipts linked
1

“Your churn isn't a pricing problem, it's a week-one silence problem.” Break down the 7-day window.

@retentionrae · 1.4k likes@saasteardowns · 890
2

Post the activation metric ladder: signup, first value, habit. Ask which rung readers lose people on.

@pmfnotes · 2.1k likes
3

Contrarian: product tours increase time to value. Show the skip-rate replies as receipts.

@onboardinglab · 760
report ready · 214 finds filed·you read it with coffee and pickexample run · fictional data

Content ideas are a volume game, and the volume part is annoying. This agent runs 8-12 standing X searches every night, files what your audience actually talks about into your workspace database, and writes you an ideas report with the receipts linked. You read it with coffee and pick.

X search + threadsWorkspace Postgres
01// install

One paste into your agent.
Your first ideas report, minutes later.

works with any agent+ 8 more
Install superagnt's Content Ideas Engine skill for me. It's the audience-radar plugin from the github.com/superagnt/plugins marketplace, plus the superagnt MCP server (on Claude it's a connector I add in my browser; elsewhere I sign in through my browser). The setup guide for each kind of agent is at https://www.superagnt.com/install?skill=audience-radar
or follow the detailed setup instructions

» installs the audience-radar skill (read every file first, or the raw SKILL.md) and enables data:x, database

» not at your machine? send the install to your inbox and run it at your desk.

02// a real run

One sweep and its report, Nov 3, 2026

The niche: Claude Code and AI for marketing. Ten standing searches with like floors, run on a superagnt test workspace at 18:12 UTC, each asking for the full payload, since long posts run to 9,000 characters. 201 posts back, no reposts, 191 different posts, 44 of them from the last 7 days. 10 credits, 32 seconds.

the ask

“Run the Content Ideas Engine skill for my niche, Claude Code and AI for marketing, and find me 5 post ideas from this week's conversations.”

  1. 01One of the ten standing searches, top posts with 20 likes or more2.6 s wall time
    → calldata_x_discover
    {
      "action": "search",
      "query": "\"claude code\" seo",
      "section": "top",
      "min_likes": 20,
      "limit": 20,
      "response_format": "raw"
    }
    ← result
    {
      "results": [
        {
          "tweet_id": "2110180038637408357",
          "creation_date": "Wed Oct 14 01:25:01 +0000 2026",
          "text": "530 people asked for the system prompt I use to run SEO + AEO inside Claude Code and Cursor.\n\nInstead of gatekeeping it in DMs, I just open-…",
          "favorite_count": 1602,
          "reply_count": 129,
          "retweet": false,
          "user": {
            "username": "illyism"
          }
        },
        {
          "tweet_id": "2114985523848663518",
          "creation_date": "Tue Oct 27 07:40:18 +0000 2026",
          "text": "25 SEO SKILLS. 18 SPECIALIST AGENTS. ONE OPEN-SOURCE TOOL.\n\nClaude SEO turns Claude Code into an AI-powered SEO analysis system that can aud…",
          "favorite_count": 91,
          "reply_count": 32,
          "retweet": false,
          "user": {
            "username": "BIGMayrr"
          }
        }
      ]
    }

    » Trimmed: 18 more posts of 20, the next-page token, and the fields this step doesn't read (views, media, the author's profile)

  2. 02File the search's posts, keyed on the post id431 ms wall time
    → callagnt_db_write
    {
      "action": "upsert",
      "table": "radar_posts",
      "on_conflict": "post_id",
      "rows": [
        {
          "post_id": "2110180038637408357",
          "query_id": 3,
          "author": "illyism",
          "text": "530 people asked for the system prompt I use to run SEO + AEO inside Claude Code and Cursor.…",
          "likes": 1602,
          "replies": 129,
          "url": "https://x.com/illyism/status/2110180038637408357",
          "posted_at": "2026-10-14T01:25:01.000Z"
        }
      ]
    }
    ← result
    {
      "success": true,
      "affected_row_count": 20
    }

    » Trimmed: 19 more rows of 20, the rest of each post's text, and a note on returning rows

Every post went into radar_posts keyed on its post id. The same ten searches had run twice earlier that day, so 121 of the 191 posts were already there and were updated, not filed twice; 70 were new.

SearchLike floorPosts backWall time
"claude code" marketing20202.5 s
claude skills marketing10202.6 s
"claude code" seo20202.6 s
ai seo agent10202.4 s
claude cowork marketing10202.4 s
"how do I" claude marketing5202.2 s
"claude code" ads10203.3 s
ai marketing agent20202.5 s
chatgpt marketing workflow20212.4 s
claude outbound10203.2 s

The ten standing searches, top posts above each like floor, as the sweep ran them.

03// the report

Five post ideas, each with the post that earned it

Claude (Opus 5.5) in Claude Code wrote this in the shape of the skill's step 3 (what repeated, what's worth a reply, and ideas that each cite the post behind them), from the 44 posts of the last 7 days, since the prompt asked for this week's conversations. Likes are as of the sweep.

Patterns that repeated:

Worth a reply today:

Five post ideas:

  • Which free marketing skills work on live data, and which are only instructions. Source: @aladagberk on 53 free marketing skills, Nov 3, 51 likes.
  • What a skill that runs your Meta ads should never do without asking: pausing ads and cloning winners. Source: @tobiadonadon_, Oct 30, 1,619 likes.
  • Your zero-click Search Console queries, turned into this week's to-do list. Source: @ChadSE0, Oct 27, 238 likes.
  • The three agents a 50-agent marketing team needs on day one. Source: @dsqjaffa, Oct 28, 627 likes.
  • Block the scrapers, keep Google: what an agent firewall does to your AI visibility. Source: @illyism, Oct 28, 465 likes.
04// step by step

Three steps to your first ideas report.

  1. 1

    Point it at your niche once

    Tell it your niche in one sentence. It builds 8-12 standing X searches with engagement floors, saved so you can tune them anytime. That is the whole setup.

    data_x_search_get_search_search · agnt_db_insert

    your niche → its standing searchesexample
    “ai agents for ops”→
    ai agents for opsgtm automationclaude code workflowsagent handoff+7 more · min 50 likes
  2. 2

    It sweeps while you sleep

    Every night it runs every search, reads the top posts and replies, and files the new finds with links and engagement numbers. Nothing repeats and nothing gets lost.

    data_x_search_get_search_search · agnt_db_upsert

    while you sleepexample
    04:00 · sweep running

    » scanning replies on 2.1k-like thread

    » filed · 47 new posts

    » skipped · 312 already seen

  3. 3

    Wake up to the report

    By morning a short report is waiting: the patterns that repeated, the posts worth replying to, and 3-5 ready post ideas, each citing the conversation that earned it.

    agnt_db_select

    your morningexample

    “Nobody talks about the handoff between agent runs. Post the checklist.”

    ↳ cited @buildermode_k · 1.4k likes

    + 4 more ideas in the report

»After the first good run, the skill offers the always-on version: a small hosted runner on a schedule, so the work keeps happening with your laptop closed. You confirm the cadence before anything is scheduled.

05// what's inside

Read every file before you install it.

SKILL.md tells your agent to turn your niche into 8 to 12 standing X searches with like floors, run them, file every new post in your database without duplicates, and write a report where every idea cites the post that earned it. Its first two headings are setup for the agent: which tools to turn on, and the tool names on newer servers.

Plain markdown your agent reads while it runs. SKILL.md is the procedure. This is the plugin exactly as it installs from github.com/superagnt/plugins: 4 files, 6.5 KB.

06// questions

Common questions

01Is there a free content ideas generator for social media?

The free generators that rank for this search (6 of the top 9 on Google US, Nov 3, 2026) write ideas from a short description you type. This skill reads what your audience posted on X and cites it, so it runs on paid data: the 10-search sweep on this page cost 10 credits. The skill itself is free, and plans are $49, $149 or $399 a month.

02Where do the ideas come from?

From 8 to 12 standing X searches you set once: topic phrases, how-do-I questions, tool names and the accounts you rate. Each sweep files the top posts above a like floor, and every idea in the report cites the post that earned it.

03Does it work for TikTok, Instagram or LinkedIn?

This skill reads X only. For short-form video, Hook Lab studies the hooks winning on TikTok in your niche. For LinkedIn, the warm outbound skills read who engages with posts.

04Does it post for me?

No. It reads X and writes you a report. It never posts, replies or sends DMs; what you publish stays yours.

05How often does it run?

Once when you ask, or every night as a small hosted agent on a schedule you confirm first. Every scheduled run is billed, and the report only covers posts filed since the last one.

06How is this different from an AI social media post generator?

A post generator writes posts from your prompt. This skill finds what people in your niche already engage with, files it, and cites it, so each idea comes with the conversation that proves people care.