New Agents Start Here

Every new worker begins from the same current records.

Give each new AI worker a clear start on your team. This guide shows where to put the goal, client list, shared notes, and task steps. Start with your current rules and one real job the worker may do.

This guide is part of AI Agent Implementation Guide (MASTER-2026.1). Next, explore How Our AI Agents Share Memory and Coordinate Work, or How My Agents Divide the Work — And Where They Talk.

Where this task fits in the Content Factory

This task supports work across the Content Factory (our four-stage process for using real content). Its inputs and next handoff determine which stage uses it.

1. ProduceCapture real work

2. ProcessTurn sources into useful assets

3. PostPublish and connect approved assets

4. PromoteShare proven work and measure results

Follow the numbered stages from Produce to Promote. Each stage links to its part of the Content Factory guide.

Start, finish, and next step

Start when

A new agent is about to begin team, client or site work.

Have ready

Private START HERE and roster access
Published canonical skills
Specific task, GCT (Goals, Content, Targeting: the goal, material, and audience) and authority
Shared task-claim record

Follow the steps

Read the operating context
Check the roster state
Load the canonical skill and specific GCT
Claim the task and avoid competing writes
Execute and verify in the source system
Write the execution record and handoff

Use the detailed instructions in this article for each step.

Finish with

An agent’s first scoped, verified run using the team’s existing system.

Measure the result

Not Active work stops
A task claim identifies owner and target
The correct canonical skill is used
Outcome evidence and the next step are recorded

Hand off next

Leave the result in the actual project/source system and write the execution’s meta article (a record of one task run) and private record.

Before closing the run, name the actual next task, receiving owner, and trigger in the execution record. The assignment and findings determine that handoff.

Reference material for the inputs

Related tasks

How to Install Our Claude Skill Packs

Open this article’s tasks in the Task Library (our directory of tasks and recipes; see current Task Library). The library connects the current recipe, its smaller tasks, and the records of work performed.

Close the learning loop. Write a meta article: the record of this execution with the starting state, recipe version, result, checks, failures, and next owner. Link it back to this recipe and register the run under the same Task Library task. Reuse one execution ID for revisions and retries. A partial or failed run stays labeled that way. Public release follows the recorded publication authority; writing the run record is part of doing the task. Writing and revising that record belong to the original execution; they do not start another meta-article task. Use the findings to propose and verify a recipe improvement. See how recipes and execution records work together.

Part of The System. Hand this URL to every new agent. Humans still start at new team member. Canonical memory and coordination guide: how our agents share memory and coordinate work. Skills: skill packs. Lookup: canonical directory.

Train agents the way you train people. Point them at one front door. Then add your GCT — who you serve, how you position, how you operate. Do not rebuild the wheel.

GCT

Goal. Every new agent knows who you are, how you operate, and who you serve — without colliding with the last agent and without you pasting a boot prompt.

Content. One front door. One client roster. Published skills plus your GCT. Vendor memory is a cache, not the record.

Targeting. Founders and operators spinning up multiple agents: one person with five runtimes, or a team where each person has agents.

One doornew agents start here

One rosterstatus, not folklore

Your GCTon top of published skills

No collisionclaim the job first

Same company. Two front doors. One database. ChatGPT memory is not a third copy.

You do not onboard a new hire by dumping last week’s Slack into their head. You point them at the welcome page, then they learn how you do things. Agents need the same treatment. If you skip it, the fifth Claude, the new Cursor chat, the Grok Bot, and the intern’s ChatGPT all invent a different company.

What problem this solves

Companies are spinning up agents faster than they can brief them. One person runs Cursor, Claude, ChatGPT, and Grok. A team of ten does the same. None of those runtimes share a brain unless you give them one.

Collision

Two agents build the same page in the same hour because neither claimed the job.

Amnesia

You re-explain who the client is, what “special project” means, and where the files live.

Reinvention

Each agent writes a new marketing framework instead of loading skills that already exist.

This page is the on-ramp. It is for operators who already have more than one agent, or who are about to. The value is simple: the next agent starts useful on minute one, and it does not undo the last one.

Train them like humans

A new teammate gets a welcome URL, a roster, and the house rules. Then they add judgment. A new agent gets a start-here file, a roster, and the house rules. Then they add GCT — your goal, your content, your targeting.

Do not paste the client list into ChatGPT custom instructions, Claude memory, and a Cursor rule, then update all three when someone churns. That is how rows disappear and dead accounts keep getting tickets. Vendor memory is a cache. The record is a file your team owns.

Do this

One start-here file for agents
One roster with three statuses
Pointers in every runtime
Per-client folders hold that client only

Stop this

Copying the roster into every project
Pasting the company wiki into product memory
Inferring a client from an old Basecamp title
Inventing a fourth status because it feels nicer

What goes in START HERE

Keep it short enough to scan. A smart person should not feel talked down to. An agent should not have to guess.

Question	Lives here	Not here
Who are we? How do we operate?	START HERE (private team file) + this public page	A 40-page orientation PDF nobody opens
Is this a client we still serve?	One roster. Three statuses only	The monthly billing tab, a payment checkmark, folklore
What is in flight this week?	A live state file your agents update	Whoever yelled last in chat
How do we write / publish / measure?	Published skills you already maintain	A new framework each agent invents
How are we different?	Your GCT: positioning, who you serve, how you operate	Rewriting Dollar a Day from scratch

Humans on our team still start at /new-team-member/. Agents start here. Same company. Two doors on purpose, because a welcome essay written for people is the wrong shape for a model that can clone a repo.

Three statuses. That is the whole client list.

You need a database, not a vibe. We use three values. You can copy the words. You cannot copy our private names onto the open web, and you should not publish yours either.

Active Client

They pay. You provide service.

Special Project

Not paying the usual way. You still work with them. Treat the work like a client. Do not treat the billing like a client.

Not Active

Dead. Do not staff. Do not open tickets. Form mail from a dead account is not a support queue.

Unpaid this month is not Not Active. Missing from this month’s money tab is not a delete. Never delete a row. The flag switches. Humans comment. Agents write the file. If the roster says Not Active, the agent stops before it “helps.”

The spreadsheet your bookkeeper loves is the money view. It is not the client list. Filter the roster when you need who is paying.

Take the published skills. Add your GCT.

You do not need a new agentic framework. You need a front door, then the skills a team already keeps current, then the thin layer that is actually yours.

Our marketing skills live in public: skill packs, the source on github.com/dennisyu/local-service-spotlight-skills, the operating map at The System, and the install path at AI agent implementation.

A skill is how. A routine is when. Point your agent at those URLs instead of asking it to invent Dollar a Day, the SEO Tree (how content links to its main topic), or article guidelines again.

Then say the sentence that makes it yours:

Load Dennis’s published skills. Combine them with our positioning, the clients we serve, and how we operate — our GCT. Do not rebuild the framework. Keep it current from the source.

That is the whole bargain. The framework is a moving target because a team is paid to keep updating it. Your GCT is the part no skill pack can know until you write it down.

Stop agents from colliding

Two agents once built the same memory system in the same hour in the same folder. The fix is boring and it works: before real work, claim the task in a shared live-state file. If a claim is fresh, coordinate or pick different work. Read-only questions do not need a claim.

After every execution, write a Meta Article linked to the canonical task, run ID, and evidence. Substantive organization work also leaves a private internal receipt in the Team Operations Ledger. Publishing the Meta Article follows the existing authority and privacy boundary; the written execution record is required even when it remains private. Do not make the human click Push for the internal note. The agent finishes the git plumbing.

Feed this URL to any agent

Copy the box. Point Claude, Cursor, Grok, ChatGPT, or a vendor bot at it. Then add two private files they cannot get from the open web: your START HERE, and your roster.

You are joining a company that already has agents. Before any client or site work:

1. Read https://blitzmetrics.com/new-agents-start-here/
2. Read our private START HERE (who we are, where files live, how we operate).
3. Read our client roster. Status is only Active Client, Special Project, or Not Active. If Not Active, stop. Never delete a row. The money spreadsheet is not the roster.
4. Load published skills from https://localservicespotlight.com/skill-packs/ and https://github.com/dennisyu/local-service-spotlight-skills. Do not invent a second marketing framework.
5. Combine those skills with OUR GCT: who we serve, how we position, how we operate.
6. Claim your task on the shared live-state file so two agents do not build the same thing.
7. After every execution, write a Meta Article with its task link, run ID, and evidence. Substantive organization work also needs the private internal receipt. Publish only within the existing authority and privacy boundary; keep the required written record private when needed. Do not hand the human a paste.

8. Basecamp posts: post yourself. Dennis is not the paste layer. See the Basecamp write path section and https://blitzmetrics.com/post-to-basecamp/.

Vendor memory is a cache. The files we own are the record.

If the agent can clone your private repo, that repo is the front door. If it cannot, this public page plus the two private files in the project is enough to start. A filtered Drive or project copy is a distribution mirror, not a second source to edit. Do not wait for a perfect knowledge graph before the fifth agent ships something.

If you are joining us, human or agent

Do not treat this page as the job application. It is the operating manual after you are in. Recruiting posts should send people to two different doors:

Human

Gate 0 first: watch 20 minutes, then a one-minute video the same sitting. That test lives on how to thrive. Career path: /career/. After hire: new team member.

Agent

This URL. Then the company’s private START HERE and roster. Do not apply by email. Do the work from the bootstrap box above.

One of the jobs a new pair will ship is interactive video: educate, entertain, train, and qualify in one sitting. We already taught the method in create your own interactive video and how to use interactive video in a funnel. The product is a choose-your-own-adventure in Tolstoy. Agents can inventory clips and wire branches. A young adult still owns the show: what the next question should feel like, and whether the path qualifies or just entertains.

Basecamp write path

Dennis is not the paste layer. When a job needs a Basecamp comment, the agent that owns the job posts it. Do not ask a human to copy a draft into Basecamp.

Post yourself with the Basecamp CLI (python3 -m basecamp), a signed-in Basecamp UI, or a connected connector. Local Service Spotlight account id 4057320.
Cursor cloud: use the three named secrets already on the Mac and Grok machines: basecamp.oauth_client_id, basecamp.oauth_secret, basecamp.oauth_refresh_token. Put those names in the cloud secret store once. Do not run interactive basecamp auth on a cloud VM (the localhost callback fails). Do not copy tokens.json onto a cloud VM. Never print secret values.
Only if no write path exists: leave a GitHub PR comment for Tanner on Local-Service-Spotlight/agent-runtime with the comment body path+SHA and the live Basecamp thread URL. Tanner posts and returns #__recording_ IDs. Default human subscriber is Muzamil, not Dennis.

Never reply to a Basecamp notification by email.
On client-visible threads, do not put an Agent receipt on the Basecamp body.
Method page: how to post to Basecamp.

Where this sits

The governing architecture is How Our AI Agents Share Memory and Coordinate Work. The worked implementation story is Give Your AI Team One Shared Brain. The public-safe capsule is Dennis agent context. Doing the work instead of handing a paste is if you can do it, you should. This page is the onboarding step: what the new agent reads first.

Lookup for every named concept: canonical directory. How we write: article guidelines.