Agents that watch Reddit, X, HN, SEO data and trends and send you digests. Each entry has a copy-paste prompt, difficulty, setup time and known pitfalls.
Isolated cron jobs scan Reddit, check inboxes, and pull Google Search Console data, then send summaries to Telegram.
Bookmarks pile up and get forgotten. Jeston Lu built a Grok Bot that scrapes saved X posts every 6 hours and auto-categorizes them into a Notion database so the archive stays searchable.
The plaque in Super Mario 64's castle courtyard is a 32x32-pixel texture fans have argued about since the 1990s. No upscaler can read it. Claude built a model of how a 1996 artist likely made it (font, canvas, resize, blur, stone), searched fonts, sizes and words on a GPU, and tested the method on fake plaques with known text. Its best reading: "Eternal Star / We hope so" in Palatino Bold Italic. Code and a 40-page report are public.
A native, clearly labelled card in the directory results.
Become the featured sponsor →Claude Code sends a pile of videos, PDFs and articles through NotebookLM and turns the output into linked Obsidian notes covering scenarios, actors and probabilities.
Same research-and-save task given to three agents. Comet used the existing in-browser Notion session and created the note in one go.
A free Claude Code skill researches a topic on the web and builds about 12 two-minute audio lessons with two hosts, a one-screen card per lesson and links to every source. A checker refuses lessons that cite pages it didn't read.
Phil Metzger asked Grok Bot to, every night after 2am, read all posts and replies by @drphiltill for the prior 24 hours, note topics, run a literature review on anything important, and tell a “chief of staff” bot whether to brief corrections, missing material, or better arguments. He reports one highly valuable correction so far; the prompt screenshot is in the post.
A user moved his bots' memory into an Obsidian vault so switching agents doesn't mean starting over. Grok Bot, Meta Muse, ChatGPT Dots and Hermes bots keep memory, logs and workflows in the same plain files he owns. Each bot reads the relevant files before answering and saves new information there (his health bot updates the health folder), and dated logs keep the history. A new agent gets pointed at the same vault with the same instructions.
A hobbyist used Codex to build tools that scan public NASA Kepler and TESS data for stars that dim on a schedule. It flagged an uncatalogued dip every 13.77 days around KIC 5689121, a small red star about 970 light-years away. It sits 7.1 arcseconds from a much brighter star, which is probably why it was missed. If it's real, it's a planet about 2.5 to 3 times Earth's size, and he published all the evidence online.
A founder with stage 4 ALK+ lung cancer wanted a trial for a new drug but not if it meant flying across the country again and again. In September the trial added a site at the University of Cincinnati, about an hour from his home, and his Grok Bot caught the change and flagged an eligibility question for his oncologist. It now checks every morning and tracks more than 25 options in Notion.
An open-source agent skill that turns Claude Code, Codex or another agent into a careful family-history researcher for the former USSR and the Russian Empire. It works through church books, civil, military, deportation and repression records, keeps a search log and registers of people and claims, matches each event to the borders of its time, and labels every claim as a hypothesis, single-source, corroborated or contradicted instead of guessing.
The agent publishes static sites to surge.sh as needed. The main use is a personal news and trends dashboard.
Every day at 5pm it sends the top posts from chosen subreddits, asks whether you liked them, and saves your preferences as rules.
Every Sunday it posts the week's tech earnings calendar. You pick companies, it schedules a job per report, and posts beat/miss, EPS, guidance and AI highlights.
Before coding anything, it scans GitHub, HN, npm, PyPI and Product Hunt for competitors and stops if the space is crowded.
Drop URLs, tweets, YouTube links or PDFs in chat to ingest them, then ask questions and get ranked excerpts with sources. Other workflows query it too.
An amateur used Claude Code to search public data and flagged what may be an undiscovered planet. It reached 301 points on HN.
Paper trades on prediction markets with backtesting, strategy analysis and daily P&L reports. No real money.
An MCP server gets the top 10 niche videos with engagement metrics and thumbnails, then hands off to one subagent that writes a research report to disk.
An agent skill makes the coding agent answer complex questions as a single offline HTML page with diagrams. The author says the model writes about 1/8 of the tokens it would need to hand-write the HTML.
Ask a biology question, like how two drugs act differently at a dopamine synapse, and a Claude Code skill builds a standalone interactive page with real structures, a toggleable mechanism and a citation for every number.
A tutorial agent picks a peer set for a public company, pulls SEC filings, news and market signals, extracts competitive moves into a table, and writes a briefing with a citation for every claim.
Every weekday at 8am an n8n workflow searches the day's AI news, scores each article for materiality at about $0.00004 apiece, and posts one cited brief per topic to Slack. Topics with no real news get no post; a morning costs 1-4 cents.
An n8n workflow collects public Telegram channel posts into a sheet for a news tracker, matching rows by post ID so a failed import can be retried without duplicates: importing 34 posts twice left 34 rows.
A dedicated Grok Bot checks Eventbrite, Luma, Meetup, LinkedIn and similar listings for founder events in Dubai, ranks them by the author’s preferences and calendar availability, and sends a shortlist for confirmation. The author credits it with getting them to Startup Valley and suggests the same pattern for cheap flights, open roles, and promotions.
Grok Bot sends a morning rundown of the most relevant X mentions, who is talking about the author’s apps, and important posts on topics of interest (local AI). The author published a read-only template.
After native X tools landed for Grok Bot, one user has it scan X every morning for Tesla and SpaceX posts, draft a digest, and fact-check material before they reply.
A user had Grok Bot research a topic, catalog the findings, make sure nothing is duplicated, and send the results every morning — running in the cloud with their own computer off.
A user had Grok Bot build and run a daily midterm polling scorecard that arrives every morning with charts and footnotes.
Tag your bot under any post on X and your Grok Bot picks up the job — a digest for a long thread, or a price-drop watch for a product post.
A US real estate investor shares a Grok Bot template built for their own underwriting: paste an address or a Zillow, Redfin or Realtor link and it scores the house for rent-by-the-room, long-term, mid-term and short-term rental, flip and BRRRR, with a short scorecard saying which fits.
A user shares his top Grok Bot use: every morning it sends him a deep dive into his day, with research on all the people he's meeting and a full report on his life and business.
Would you rather listen to your day than read it? His main Grok Bot has its own podcast show, with hosts, sections and a jingle, that sums up everything that happened during the day and the week.
Does your feed throw a new AI tool at you every day? A private equity professional pastes each new tool's link into his Hermes Agent and asks "useful for me or not". It reads the site and answers against what he actually runs, checking 3 things: how many agents would use it, whether it reaches the app he lives in all day (at work, his email only works through the desktop app), and whether his agent already does it. The last one he asked about got "not now, maybe later". He also had it list every change he'd made to it, and found some fixes would be wiped by the next update.
He asked his ChatGPT Dots assistant to backtest trading strategies taken from popular YouTube trading channels. None of them came out with a profit factor above 1, which means none made more than it lost in the test.