Most investors run Claude one prompt at a time. Day 12 stacks six skills into a pipeline that runs while you sleep. It scans the market, underwrites the shortlist, tracks the portfolio, chases the rent, and sources maintenance quotes. A 15-bullet briefing lands in your Discord before you open your laptop. The work a small back office does, on infrastructure Anthropic hosts for roughly fifteen dollars a month.
Most property investors using Claude are running it the way they ran the search bar before it. One question at a time. A market scan in the morning, a deal analysis in the afternoon, the same prompt re-typed every time a new launch lands. The work is good. The leverage is missing.
The leverage is in the second move. Convert each repeated prompt into a skill with a system instruction and a folder of inputs. Convert the skills into a pipeline that runs on a schedule. Convert the pipeline into a single 15-bullet briefing that lands in the same place every morning. The investor stops asking Claude to do things and starts reading what Claude already did.
Anthropic ships the infrastructure that makes this possible. Managed Agents run on a sandbox Anthropic hosts, schedule themselves through the Sessions API, and post output anywhere a webhook can reach. The cost is roughly fifteen dollars a month for a pipeline a small property office would charge five thousand to run.
One prompt at a time is a search bar. Six skills on a schedule is a back office.
A three-layer pipeline that runs every morning at 6am, unattended. Layer one is the infrastructure. Layer two is the six skills. Layer three is the schedule and the delivery.
Layer one is a single Managed Agent on platform.claude.com. Anthropic hosts the sandbox, the file system, the runtime. You define what the agent is allowed to do and where its inputs live.
Layer two is six skills the agent runs sequentially: Deal Scout, Investment Analyst, Portfolio Manager, Rent Recovery, Property Operations, and Morning Briefing. Each skill has its own system prompt and its own input folder.
Layer three is a daily 6am trigger that opens a session and runs the agent end to end. Morning Briefing posts the compiled output to a Discord webhook before you open your laptop.
A Managed Agent created on platform.claude.com with the agent_toolset_20260401, a workspace mounted at /workspace, and a system prompt that names the six skills and their inputs.
No cron job on a Mac that has to stay awake. No server you maintain. No environment variables that drift.
Deal Scout filters the launches. Investment Analyst underwrites the shortlist. Portfolio Manager tracks the holdings. Rent Recovery chases the overdue. Property Operations sources the maintenance quotes. Morning Briefing compiles the lot into fifteen bullets.
Each skill has its own system prompt, its own input folder under /workspace, and its own output folder. The order matters, because Morning Briefing reads everything the other five wrote.
A cron job on any always-on machine, or the Claude Code Desktop /loop command, fires a Sessions API call at 6am every day. The agent wakes up, runs the six skills, posts the briefing to the configured Discord webhook, and closes the session.
Session-time pricing means the cost is metered in minutes, not subscription tiers. A typical 30-minute morning run is about $1.20 a month in runtime plus token usage on top.
An Anthropic API key from platform.claude.com. The Pro and Max subscriptions are for Claude.ai. They do not include Managed Agents access. Managed Agents bills against the API key, separately.
A workspace folder structure on your machine: /watchlists/, /holdings/, /inbox/, /ledgers/, /suppliers/, /memory/, /outputs/. The folders can be empty at first; you fill them as the pipeline runs.
A Discord server with a webhook configured, or any other endpoint that accepts an HTTP POST. The reference uses Discord because the webhook setup is one menu deep and free.
Roughly an hour for the first run end to end. After that, the pipeline is unattended.
A Claude Max ($100/mo) subscription alone does not give you Managed Agents. Managed Agents is a platform.claude.com feature billed against an API key with usage-based pricing. You can run the pipeline on a Pro subscription for Claude.ai chat in parallel; the API key is the part that matters for this guide. Spend on the pipeline itself is roughly $5-10/month in tokens plus $1.20/month in runtime at a 30-minute morning session.
Managed Agents is the Anthropic-hosted runtime for long-running, sandboxed agents. You define the agent and its tools once. Anthropic owns the container, the file system, the runtime memory, and the lifecycle. You never SSH into anything.
Route A. Claude Code TUI. Open Claude Code, paste the scaffold prompt below, and let Claude create the agent and the environment for you. Faster, lower friction, fewer chances to fat-finger a header.
Route B. Manual API. Two curl requests, one to create the agent and one to create the environment. Slower, but the only path if you want the agent definition checked into source control.
The environment is the persistent sandbox. It keeps your /workspace files alive between sessions. Without it, the agent starts from a blank file system every morning, the six skills cannot read each other's outputs, and the Morning Briefing has nothing to compile. Create the environment once. Reuse the environment_id every day.
Each skill below is a self-contained system prompt. Save each one as its own markdown file in /workspace/skills/, name the files in the same order the agent should run them, and reference them by filename in the Property Pipeline agent's system prompt. The skill prompts are the work. Everything else is plumbing.
One giant prompt that tries to do all six jobs in a single pass will compress, hallucinate, or lose track of the inputs. Six separate skills run sequentially keep each task in its own context window with its own inputs and its own output file. Morning Briefing only works because the other five wrote their outputs to disk first. Splitting the work is the architectural decision that makes the pipeline reliable.
Two scheduling options. Pick one. The first is production-grade and runs from any always-on machine. The second is one line of Claude Code Desktop and runs from your laptop, as long as the laptop stays awake.
Option A is unattended forever. You set the cron once and the pipeline runs whether your laptop is open or not. Option B is one line, no infrastructure, but it stops the moment your laptop sleeps or Claude Code Desktop quits. If you travel or your machine sleeps overnight, use Option A. If you keep one always-on Mac and want to be running tomorrow, Option B gets you there faster.
The pipeline writes the morning briefing to disk for the audit trail. It also posts the briefing to wherever you actually read messages at 6:01am. The reference uses Discord because the webhook setup is free, one menu deep, and accepts a plain HTTP POST. Any other channel that accepts a webhook substitutes one to one.
One. Open the Discord server you want the briefing in. Open Server Settings, then Integrations, then Webhooks.
Two. Create a new webhook. Name it "Property Pipeline" and pick the channel it should post to. Copy the webhook URL.
Three. Export the URL as an environment variable on whatever machine runs the cron: export DISCORD_WEBHOOK="https://discord.com/api/webhooks/..."
Four. Test it once by hand. The Morning Briefing skill posts the same way the test command does.
If you do not use Discord, swap the webhook URL for any HTTP POST endpoint. Telegram via a bot, Slack via an incoming webhook, email via a Mailgun or Resend POST, or a private iOS shortcut webhook. The Morning Briefing skill does not care what is on the other end of the POST.
Managed Agents pricing is metered, not subscription. You pay for the API tokens the six skills consume, plus a small runtime fee for the minutes the session is active. Real-world cost for a six-skill pipeline running once a day is in the low double digits per month.
Per month. Six skills, one run a day, sonnet-4-6 pricing. Heavier on Investment Analyst days when the shortlist runs long.
Per month. Roughly 30 minutes of active session per morning at $0.08 per active hour.
Discord webhooks are free. Same for Telegram bots. Email and SMS substitutes have their own metered pricing.
The pipeline is good at filtering, drafting, and compiling. It is not good at deciding. The five things below are where the investor still has to do the work.
Deal Scout filters on the buy box you wrote. If the buy box is loose or out of date, the shortlist will be longer than it should be. Tighten the buy box once a quarter. The pipeline is only as sharp as the criteria it scores against.
Investment Analyst writes a one-page underwrite from the inputs in front of it. It does not visit the site. It does not read the SPA the way a real-estate lawyer reads it. The memo is the screening artefact that decides whether to spend the diligence budget, not the diligence itself.
Rent Recovery and Property Operations write drafts. The investor reads, edits, and sends. The pipeline is deliberately built so that nothing the agent writes leaves the investor's machine without a human review step. Skipping that step is how a tier-3 escalation goes out to the wrong tenant.
MEMORY.md and the per-skill memory files (comparables.md, decisions.md, maintenance-history.md) are append-only by default. Stale entries make the agent slower and the briefings worse. Trim the memory files monthly. Archive what you no longer want the agent to use.
If the Discord webhook fails, the briefing still lives on disk. The agent does not retry beyond once. Build a habit of opening /workspace/outputs/morning-briefing/ as a backup. The infrastructure is reliable; the integrations are integrations.
Earlier days in the series each build one capability. Day 01 underwrites a single property. Day 09 protects a single off-plan unit. Day 11 puts a comp desk at the launch room door. Day 12 takes the skills you have already learned and runs them on a schedule, in parallel, without you in the loop.
A skilled investor running prompts one at a time is faster than the market. An investor whose skills run on a 6am schedule is operating on a different surface area entirely. The compounding edge is not the prompts. It is the fact that the prompts run every day, whether you remember to or not, against every new launch and every overdue tenant and every open maintenance request, before you make your first decision of the day.
By 6:01am every morning, a fifteen-bullet briefing has landed in your Discord. The top section names the three things that matter most today. Below it sits every new deal worth a second look, every underwrite ready to read, every portfolio action due in the next fortnight, and every rent and operations draft awaiting your send.
The day starts not with you asking Claude what to do, but with you reviewing what Claude already did. That is the architecture shift Day 12 ships.
Every prompt in this 30-day series is one capability. FourthspaceOS bundles all of them into a single product: underwriting, comps, market research, deal sourcing, portfolio tracking, and investor reporting. The product runs natively on the same Anthropic agents and plugins you are learning to use this month.
Waitlist members get founding pricing, early access, and one new prompt delivered each day for the next 30 days.
Founding pricing locks in for waitlist members.
One email a day for 30 days. No spam. Unsubscribe whenever.
Most investors stop at the first useful prompt. Send them this page. The leverage from a single skill is real. The leverage from six skills on a schedule is a different category of edge.
Find me on Instagram ↗You do not need all six skills running on day one. Pick the one that hurts most. Wire Deal Scout if your inbox is the bottleneck. Wire Rent Recovery if cashflow is. Get one skill running on a schedule and the others compound from there.
Jump to the six skills ↑