Cold email that improves itself.

What cold.md is

Most outreach tools keep your targeting, voice and offer inside their own settings, where nobody can diff it and nothing else can read it. cold.md puts it in one markdown file at the root of your repo, in eight sections, in this order:

• ## Identity who's sending
• ## Audience who receives it, and who never should
• ## Value the offer, in one sentence
• ## Voice dos and don'ts, obeyed word for word
• ## Proof the only facts the agent may cite
• ## Sequence opener, bump and breakup slots
• ## Objections your answers to the usual pushback
• ## Banned phrases that never appear

The rules are hard rules. A banned phrase is never written, voice rules are followed word for word, and only facts listed under Proof can be cited. The spec is open (CC-BY-4.0), and the plugin is MIT. FoxReach is the sending layer it calls.

Prerequisites

Install

Inside a Claude Code session:

/plugin marketplace add foxreach/cold-md
/plugin install cold-md@cold-md

Or from a terminal, which does the same through the claude CLI:

curl -fsSL https://cold.md/install | bash

Then connect FoxReach in the shell where Claude Code runs:

pip install foxreach-cli
export FOXREACH_API_KEY="otr_your_key"

In your project, run /cold init. It asks nine short questions and writes .cold/config.json. The step-by-step guide covers every option.

The loop

One command per step. Each one reads and writes files you can review and commit.

1. /cold init Writes .cold/config.json, checks your FoxReach key and lists your connected mailboxes.
2. /cold icp https://your-site.com Builds icp.md from your site and checks it with web research.
3. /cold leads --finder --count 200 Searches Lead Finder (or imports your CSV), scores every lead against icp.md, and verifies emails.
4. /cold experiment Picks one thing to test and writes the protocol before anything sends.
5. /cold draft Writes the control and the variant, both checked against your cold.md.
6. /cold send Runs pre-flight checks, creates the campaign, and asks before it starts.
7. /cold triage Categorizes replies and drafts answers. Nothing is sent without your yes.
8. /cold learn Tests the variants against each other and proposes an edit to cold.md.

What runs on FoxReach

The sending skills call the public FoxReach CLI and API. These are the pieces they use.

• Lead Finder. cold-leads turns icp.md into Lead Finder filters, shows the free match count, and asks you to approve the reveal cost before any work email is revealed.
• Email verification. Every CSV lead is verified before import. Bad addresses never reach a campaign, and only definite good or bad verdicts use a credit.
• Provider matching. cold-send reads how your mailboxes and leads split across Google and Microsoft, and turns provider matching on when both sides span both.
• Mailbox health. Any mailbox whose warmup mail lands in the inbox less often than your floor is left out of the campaign before it starts.
• Reply categories. cold-learn reads interested, not-interested and out-of-office counts per A/B variant, the numbers its test runs on.
• Your inbox. cold-triage reads unread replies, sets each category, and drafts answers from the Objections section of your cold.md.

Tests that edit the file

/cold experiment picks one variable to test: subject line, opener, call to action, cadence or tone. You add the variant to the campaign step in FoxReach with Add A/B Test, using the copy the skill hands you (the API reads variants but does not create them yet). See A/B testing for the editor.

After the test window, /cold learn compares interested-reply rates with a two-proportion z-test. If one variant wins, it writes the change to your cold.md as a patch and shows you the numbers. You apply it or reject it. After three approved changes in a row, it can apply them on its own; one rejection turns that off. The full write-up explains the loop.

Troubleshooting