Collection · 13 power moves

Power moves: the setups most people miss.

Level up and Advanced entries that use several bots, run overnight, or use websites for you when there is no app. Ranked by value. Check the "on its own" label before you copy one.

Code & SoftwareAdvancedWell documented9/10

Engineer bots that manage 200+ Cursor cloud agents with a 30-minute PR board

Five domain-owner bots each launch Cursor cloud agents with the engineer's skills, watch transcripts and screenshots, and push back when proof is missing. Every 30 minutes they sweep a shared Notion PR database for bot comments, failing CI, and merge conflicts. Only high-confidence, low-blast-radius PRs auto-merge; the rest wait for human review.

Acts after you say OKAuto-merges low-risk PRs only

Several botsWorks overnightHands work to coding agentsRuns on a schedule

How it runs

One bot per domain with its own memory.
Bot starts a cloud agent with your skills and states the proof it expects (e.g. before/after screenshot).
Every 30 min: check Bugbot/security findings, CI, conflicts; send fixes back to the agent.
Clean PRs get a review run; auto-merge only if confidence is high and blast radius is low.

Reported resultAuthor went from managing 15 cloud agents by hand to a fleet managing 200+ at once; guide also says a teammate shipped 2,000+ PRs in a month.

Steps, sources, proof

Teams of BotsLevel upWell documented9/10

One Chief of Staff bot as the single front door to a specialist team

You talk to one coordinator bot. It routes asks to specialists (inbox, calendar, intel) and consolidates their reports into one stream, so three bots pinging you becomes one brief. The official guide publishes the exact role prompts, each with a quiet-on-noop rule and draft-then-approve on any send or calendar write.

Writes drafts, you send

Several botsStarts switched off

How it runs

Bot Boss prompt (official): "You are Bot Boss, the single front door for my bot team: route my asks and the specialists' reports through one stream (Inbox Manager, Calendar EA, Intel Scout) so I never talk to them directly; never invent urgency, never invent…

Reported resultAuthor says he went from chat-assisted work to orchestrating dozens of bots; no time-saved number given.

Steps, sources, proof

Teams of BotsLevel upWell documented9/10

Overnight loop that ends at a morning review list

Let routines do research, drafting, enrichment, and staging overnight, and stop at a list you review in the morning. The official docs end every starter prompt at a review point, and the SDR guide stages enrolled-but-unsent outreach. This avoids the failure mode where a 3 a.m. approval card expires before anyone sees it.

Writes drafts, you send

Works overnight

How it runs

Official Sales Outbound starter: "Research the 25 accounts in this CRM view... draft email and LinkedIn outreach... Skip anyone already in an active sequence. Return a review list; do not send or enroll anyone." Then: "create a nightly research routine that…

Reported resultSDR guide: day starts with work to review instead of an empty chat; one backlog pass fixed 24 mis-sequenced contacts. No time figure.

Steps, sources, proof

Your BusinessAdvancedWell documented8/10

SDR outbound with CLEAR vs FLAG intake and staged, unsent enrollments

Cursor agents build research packets and first-touch drafts; Grok Bot checks each contact against the sequencer, email tool, and CRM. CLEAR contacts are enrolled unsent; FLAG contacts wait for a human. A chief-of-staff bot owns the morning, a sheet bot prevents double-working contacts, and a LinkedIn bot sends blank requests after checking caps.

Writes drafts, you send

Several bots

How it runs

Rules from the guide: never auto-send; warm history beats a cold template; footprint beats title; blank LinkedIn requests, flag if the profile URL can't be verified; repeated feedback becomes a skill.

Reported resultOne morning: 6 contacts left alone (already active), 34 flagged for review, 1 sent; separately 24 mis-sequenced contacts fixed in one pass.

Steps, sources, proof

Teams of BotsAdvancedSome evidence8/10

An ops bot that holds daily 1:1s with your other bots and runs postmortems

One non-coding bot meets each worker bot every morning to review the playbook and surface blockers. When a bot makes a mistake, it is sent to the ops bot for root-cause analysis; the ops bot updates the shared playbook and announces the change to the team. The same bot onboards new bots with the team rules.

Just reads and reports

Several botsRuns on a schedule

How it runs

Create an ops bot (no code work). Routine: daily 5 a.m. 1:1 with each bot to restate key workflows. On any mistake: "Find [ops bot] for root-cause analysis and a postmortem"; ops bot edits the playbook and broadcasts the change.

Reported resultAuthor says bots rarely forget complex workflows even after many weeks.

Steps, sources, proof

Your BusinessLevel upSome evidence8/10

Trade-business office work across portals with no API

Owners use a bot as an office manager that works in the actual web portals: permits, inspection booking forms with CAPTCHA, e-signature subcontracts, client work-order portals, ServiceTitan, and a freight TMS. The reported work includes catching an insurance-certificate discrepancy and building proposals staged for approval.

Acts after you say OK

Uses websites for you

How it runs

Start with one portal you log into daily. Sign in yourself on the bot's computer, have it do one job while you watch, save it as a skill, and keep e-sign sends and bookings behind approval.

Reported resultPlumbing: 5-6 urgent work orders a day booked, about 10 minutes each (per Agents In Use summary). Roofing: nine back-office jobs in two days. Freight: hundreds of loads tracked, 100+ drivers texted, PODs uploaded to the TMS in a few hours.

Steps, sources, proof

Your BusinessLevel upSome evidence8/10

Two sales bots that search for events every two hours and leave drafts

An AV production company runs a research bot and a sales bot that search Southern California events every two hours, pull four new leads, and write an outreach email for each into the owner's drafts folder. A chief-of-staff bot turns RFP attachments into estimates in a quoting system the bot also built.

Writes drafts, you send

Several botsRuns on a schedule

How it runs

Give the bot ~20 planner terms (conference, gala, fundraiser), a venue list, and a 14-month window; routine every 2 hours: 4 new leads, one draft each, saved to drafts; owner reviews and says send.

Reported resultPipeline up 140% in about a month; 30+ RFP replies in three weeks; 11 events booked, about 15 quotes pending; estimates in 30 seconds vs 1-2 hours; 2-3 quotes a week to 10-12.

Steps, sources, proof

Money & ShoppingLevel upSome evidence8/10

Two-bot subscription audit with a gated retention negotiator

An email bot builds a table of every active subscription from receipts: cost, billing date, how to cancel. A second bot takes one target at a time, waits for the owner to send "run it", opens the vendor's support chat, asks for a retention offer, and stops at the final cancel button. It is barred from accepting deals or confirming charges.

Acts after you say OK

Several botsUses websites for you

How it runs

Bot 1: "Find every subscription I'm still paying for, what it costs, when it bills, and how to cancel each one." Bot 2 rules: log offers; never accept a deal, confirm a charge, or do anything binding without asking; start only on "run it"; stop at the cancel…

Reported resultAbout 13 subscriptions found, three near $300/month combined; on the first target it found the bill was already $39.20, not $49.

Steps, sources, proof

Code & SoftwareLevel upWell documented7/10

Bug report to reproduction pack in staging

The bot reads a bug report, reproduces it in staging with a test account, and returns exact steps, expected vs actual behavior, screenshots or video, browser details, and console or network notes. One user's bot saw a bug complaint in an Instagram inbox, handed it to a QA bot that tested the funnel, and had a writer bot draft a reply.

Just reads and reports

Several botsRuns when something happens

How it runs

Official: "Read this bug report and reproduce it in staging using a fresh test account. Return exact steps, expected and actual behavior, screenshots, browser and OS details, relevant console or network notes, and a minimal test case if possible. Do not use…

Reported resultNo metrics; the Instagram case found no bug and drafted a reply.

Steps, sources, proof

Create & MakeAdvancedSome evidence7/10

Read the codebase, generate every placeholder asset, wire it back in

A game developer had the bot read his codebase to learn what each placeholder was, write a prompt per asset in his own art tool's web page, generate the images, crop them to transparent PNGs, and wire them into the game. He then had it build a playtesting skill and an itch.io upload on each GitHub push.

Acts on its ownInside your own project

Uses websites for youHands work to coding agents

How it runs

"Visit [art tool URL] and generate art to replace every image placeholder in this repo. Read the code to understand each asset, write one prompt per asset, export transparent PNGs, and wire them in. Show me a contact sheet before committing."

Reported result74 cleaned card assets generated and integrated in about 2 hours.

Steps, sources, proof

Create & MakeLevel upSome evidence7/10

Timestamp in, HD clip out, then researcher and writer bots prep the post

Send a YouTube link with start and end times. A clipper bot returns the HD clip in chat a few minutes later and passes it to a researcher bot for quotes and a writer bot for the caption. The user set it up by signing in to Google on the bot's computer and asking.

Writes drafts, you send

Several bots

How it runs

"Cut [YouTube URL] from 16:43 to 18:17, send me the HD clip, then hand it to @Researcher for the best quotes and @Writer for a caption draft. Don't post anything."

Reported resultClip ready in a few minutes; no time-saved figure.

Steps, sources, proof

Create & MakeAdvancedSome evidence6/10

Three-bot lab: photos to simulator to experiments to 3D print

An MIT professor set a chief-of-staff bot, a physics-experimenter bot, and a 3D-printing bot on a materials problem. From four reference photos they inferred design principles, wrote an interactive simulator, ran experiments, wrote a LaTeX report, and printed the best designs.

Acts on its own

Several bots

How it runs

Roles: Chief of Staff (coordinates, moves files), Experimenter (simulator, experiments, report), Printer (slice, generate code, send job, monitor).

Reported resultDirectory summary: 47 experiments run and best designs printed on a Bambu Lab H2D.

Steps, sources, proof

Teams of BotsLevel upSingle claim6/10

One prompt that interviews you and spins up its own bot team

Create one chief-of-staff bot and paste a setup prompt. It interviews you, writes its own operating manual, specs and launches four teammates, and schedules routines. The author's rule is that nothing sends, posts, or ships without his approval.

Acts after you say OK

Several bots

How it runs

See the author's setup prompt in the first reply of the linked post. Keep new routines disabled until you review each one.

Reported resultFour teammates and six routines created from one prompt.

Steps, sources, proof

+Other collections

Open in the explorer.