For agents

Hand gtme to Claude Code or any agent with a shell in one paste line, and the rules it follows before anything spends or sends

What you'll learn

You'd rather have Claude Code or another agent do the typing, and you want to know what it will and won't do on its own.

  • the paste line and the plugin
  • the rules the agent follows about spending and sending
  • what the binary enforces and what only the rules do

gtme is a single binary that runs a go-to-market campaign written as a YAML file and prints a receipt of what each run did and what it cost.

This page needs an agent. Use Claude Code for the plugin, or any agent that can run a shell for the paste line.

This page spends nothing. The paste line and the plugin install cost $0. Each start page states what its run spends right before the command, and the agent reads that line too.

Paste this line into your agent:

Follow gtme.run/start.md

The line points the agent at START.md, the whole instruction set in one file, and you can open it before you paste. The agent installs gtme, then walks one section to a receipt. A section is one path from nothing to a receipt. There are four, from See it run to Add a vendor, the same pages this collection walks by hand.

We recommend See it run first, because it costs nothing and proves the install. After that, pick the page that matches what you have: a CSV, vendor keys, or a vendor gtme doesn't ship.

The plugin

In Claude Code, the plugin is optional and adds four slash commands. Paste the line first, then add the plugin if you want the commands. You can do both.

/plugin marketplace add gtme-run/gtme
/plugin install gtme@gtme-run

Claude Code picks a skill from what you ask for, or you can call one by name:

SkillWhat it does
/gtme:create-pipelineTurns an outreach goal, in your words, into a pipeline, the YAML file that lists a campaign's steps
/gtme:run-pipelineRehearses, runs, or ships a pipeline or bundle. Shipping is the armed run, and on a live target that command is yours
/gtme:create-adapterWrites an adapter for a vendor or data source gtme doesn't ship, or a new entity type, meaning a kind of record such as posts or jobs
/gtme:analyzeAnswers what a run did, what it cost, why a record dropped, and where each fact came from

The skills hold procedure only. They take their facts from the installed binary, from START.md, and from the pattern bundles in the gtme repository, so they track the version you have.

The rules the agent follows

Nothing sends until you run it, and nothing spends until you've read what it costs. START.md ends with five rules the agent reads before its first command. The run-pipeline skill restates the arm gate and the key rule. The create skills stop at plan or simulate and hand off to it, and analyze only reads, so it can't spend or send.

  • Never arm a live target. Arming means running gtme run without --simulate or --dry-run, so it runs for real. A live target is a vendor the pipeline sends to, such as an Instantly campaign, and that run is yours after you've read the dry-run receipt. A local target, such as a CSV on your machine, is the agent's to run after you've read the line that says what it spends.
  • Never handle a key. gtme secret set KEY prompts you, and you type the key. The agent never puts one on a command line, in a file, or in the chat.
  • Announce spend first. The agent climbs the gate ladder in order. gtme plan and --simulate spend $0. A dry-run spends on vendor credits and model tokens and holds only the send. So the agent tells you the cost before any command that spends, and waits for you when the target is live.
  • Follow the error. Each error message names its fix. The agent does the named thing and runs the same command again.
  • Stop at "done when." Each start page ends with a condition. The agent reports the receipt and waits, because the next step is yours to take.

The binary doesn't know who typed a command. The arm gate is the agent's rule plus your read of the dry-run receipt. What the binary enforces is that plan and simulate spend nothing and a dry-run sends nothing.

Where the agent looks things up

The agent reads the whole CLI and adapter surface from gtme help --agent when it needs to, and you never have to. Agent-operable by design explains why, and the help reference documents the command.

Next

Use gtme from Claude Code walks a full session from goal to receipt. If you'd rather see a receipt with your own eyes first, run See it run by hand.

  • See it run — The page that needs no keys and spends nothing, and the one we recommend the agent start with
  • Add a vendor — The last of the four start pages; an agent walks the same pages this collection walks by hand
  • Runs and receipts — Every start page ends in a receipt, which is what the agent reports back to you
  • The gate ladder — The rungs from simulate to armed; the agent climbs through dry-run and you arm a live target
  • A campaign is a folder — The run-pipeline skill runs bundles as well as single pipeline files
  • Two adapter tiers — The create-adapter skill writes adapters for vendors gtme doesn't ship
  • Plugin skills — Each plugin skill's trigger, procedure, and the commands it runs
  • Agent-operable by design — Why the binary describes its whole surface in one machine-readable document an agent reads when it needs to
  • gtme help — The help verb, including the --agent output the agent reads
  • Use gtme from Claude Code — The next page for this reader, a full session driving gtme from Claude Code