# Outreach opener

Researches a person and their company, then saves a short first email to them as a Gmail draft.

Requires Gmail, through Composio. Requires web search, through Exa. Requires web fetch.

## Install

No one has run this against real Gmail accounts through Composio yet. Its eval cases use stand-ins.

```sh
npx @parcha/agentrun template install github:Parcha-ai/agentrun-templates/templates/outreach-opener@242f29da8317fbe51d0cce4628bdb5ab61d7ccc9 --trust
```

Or open `<your AgentRun address>/#/templates/install?src=github%3AParcha-ai%2Fagentrun-templates%2Ftemplates%2Foutreach-opener%40242f29da8317fbe51d0cce4628bdb5ab61d7ccc9`. Pinned to commit 242f29da8317fbe51d0cce4628bdb5ab61d7ccc9.

## What it does

Give it a person, their company and their email address, and it researches both on the public web, then writes a short first email that opens on one recent, sourced fact about them or their company and says why you're writing. It saves the email as a Gmail draft to that address and never sends; the draft waits for your yes before it's saved. It never claims a fact it didn't read, never treats an old fact as news, and says so when it can't confirm the person works there.

- Skill `outreach-opener`: Research one person and their company on the public web and save a short first outreach email as a Gmail draft to the address given, through Composio's Gmail draft tool. Confirm the person works there, find one recent sourced fact to open on, tie it to why the owner is writing, and save one draft. Never a send. Use for "write an opener to \<person\> at \<company\>".

## What starts it

- By hand, any time: from the console, the API or the CLI.

## Inputs

| Name | Type | Required | What it is |
| --- | --- | --- | --- |
| `person` | string, 2 to 120 characters | yes | Their full name. Example: `Kim Lee`. |
| `company` | string, 2 to 160 characters | yes | Where they work. Example: `Globex`. |
| `email` | string, email | yes | The address the draft goes to. The only one it ever uses. Example: `kim@globex.example`. |
| `website` | string, uri | no | Tells the company apart from others with a similar name. Blank: the domain of their email. Example: `https://globex.example`. |
| `profile_url` | string, uri | no | A public page about them (a bio, a team page, a talk). Blank: it searches. Example: `https://globex.example/team/kim-lee`. |
| `reason` | string, 0 to 600 characters | no | Why you're writing to this person now, if you know. Blank: it works it out from what it finds and your about\_you. Example: `They're growing the retail ops team and we sell shift scheduling.`. |

### Setup, filled in once at install

| Name | Type | Required | What it is |
| --- | --- | --- | --- |
| `about_you` | string | yes | Who you are and what you offer, in a line or two: "I run sales at Shiftly; we make shift scheduling for retail chains." Every opener says why you're writing from this. |
| `sign_off` | string | no | How you sign: "Best, Mia" or "Mia Torres, Shiftly". Blank: no name, so add your own. |

## What it needs

- Harness: pi, claude-agent, codex
- Model: balanced tier, medium effort
- Tools: read, web_search, web_fetch
- Composio Gmail tools: GMAIL_CREATE_EMAIL_DRAFT. Saves one draft to the address you type in. It can't send and reads nothing in your mailbox. The address is the run's input, which no setting can fix in advance, so each draft asks you first; cc, bcc and extra recipients are always off.
- Computer: Wherever your AgentRun puts a new agent
- Keys: A model key
- Learning: on

## Sample output

Input:

```json
{
  "content": "Write an opener to Kim.",
  "person": "Kim Lee",
  "company": "Globex",
  "email": "kim@globex.example",
  "website": "https://globex.example",
  "reason": "They're growing retail operations and we sell shift scheduling."
}
```

Output:

Illustrative output, written by hand to show the shape. A measured eval run replaces it.

Draft saved to **Kim Lee** (kim@globex.example). Not sent.

It opens on Kim's new role as VP of Retail Operations ([Globex, Nov 2](https://globex.example/news/2026-11-02-kim-lee-vp-retail-operations)) and the 40 stores planned for 2027 ([RetailWire, Oct 28](https://retailwire.example/2026/10/28/globex-40-stores)).

> **Staffing 40 new Globex stores**
>
> Hi Kim,
>
> Congratulations on the new role. Leading store staffing while Globex opens 40 stores and hires about 1,500 people next year is a big first project.
>
> I run sales at Shiftly. We make shift-scheduling software for retail chains, and new stores are usually where scheduling gets hardest: new teams, no history to plan from.
>
> Would it be worth 20 minutes in the next few weeks to compare notes on how you plan to staff the openings?
>
> Best, Mia

Check before sending: the 20-minute ask, if you'd rather open without one.

## Tested

Passes 4 of 4 eval cases on pi, 3 of 4 on Claude Code and 4 of 4 on Codex. Not yet tried on the live web.

4 eval cases: injection-in-page, recent-fact-hook, same-name, stale-facts-only.

## Source

- Package: https://github.com/Parcha-ai/agentrun-templates/tree/242f29da8317fbe51d0cce4628bdb5ab61d7ccc9/templates/outreach-opener
- Publisher: Parcha-ai
- Author: Parcha Labs
- Version: 0.1.0
- License: Apache-2.0
- Category: Sales
- Page: https://marketplace.agentrun.ai/templates/outreach-opener/
