claude-cold-email
AGENTS.md3 starsChanged 6 months agoMITView on GitHub
This file documents how the cold-email skill should behave when an agent (Claude / Cursor / Codex / any MCP-aware runtime) is operating inside a project that has installed this pack. These rules are LOAD-BEARING. Skills check these on activation and adjust behavior accordingly. Three files combine to define how the agent behaves: The skill enforces the framework. SOUL + brand-config personalize the output. Never let SOUL override the framework's banned-patterns list. Before any output, agent must load both files…
1 other project uses a version of this.
Adopted by
Most recently taken up by
- cmj-hub/gtm-operator-skillsword for word, first seen Nov 1, 2026
Other versions1 project uses a version of this.
Word for word · 1
- cmj-hub/gtm-operator-skills★ 3100% same
Edited · 0
None found.
Wrote their own version · 0
None found.
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
Where files are headingNew AGENTS.md files have more rules about keeping secrets out of the code, and fewer about code style.
Of the AGENTS.md files we read, the 32,671 first seen in the last six months, against the older ones: the kind of rule that gained most share, and the kind that lost most. About files like this in general, not this one, and an estimate, since a kind is a pattern in the words.
In this repositoryOne other file it gives its agents.
- cold-emailSkill
The file
Show its rules and text
We split this AGENTS.md into the rules it gives an agent
17 rules. About 0 are common, 9 are distinctive, and 8 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈9
Distinctive · estimate
This author's own choices. Worth a look.
8
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 27
Before any output, agent must load both files (if they exist) into context.
- Very specific: line 33
If the operator has not set up brand-config.json + SOUL.md AND has not explicitly opted out (e.g.
- Distinctive: line 39
When using a public signal, quote the signal in the operator's source (LinkedIn post / job listing / press release).
- Distinctive: line 51
If a required input is missing, ask.
- Very specific: line 55
Every output runs through the self-check rubric in skills/cold-email/SKILL.md before delivery.
- Distinctive: line 61
Never connect to send APIs without explicit per-call authorization.
- Distinctive: line 65
When the skill routes to a mode, that mode owns the workflow.
- Distinctive: line 70
When running the deliverability mode against blacklist zones, respect their rate limits.
- Very specific: line 75
If an experiment log path is configured (operations.experiment_log_path in brand-config), log every cold-email draft + every nurture …
- Distinctive: line 77
every audit run with timestamp + inputs + output summary.
- Distinctive: line 82
If signal / pain / EVP / vocabulary are uncertain, flag the uncertain parts and ask for the actual data.
- Very specific: line 87
Write cold emails without a brand-config.json (refuse + route to onboarding)
- Very specific: line 88
Use banned patterns from skills/cold-email/references/banned-patterns.md even if the operator requests them — push back with the JMC …
- Very specific: line 98
If brand-config.json and SOUL.md are both missing on first invocation:
- Very specific: line 104
If yes → run /cold-email:cold-email setup (skills/cold-email/modes/setup.md)
- Distinctive: line 105
If no → minimal-mode: agent will produce framework-shaped outputs but won't personalize them.
- Distinctive: line 111
No telemetry. No outbound calls except:
≈ “Common” is an estimate: at least 1 in 5 of the 41,581 AGENTS.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
AGENTS.md — Behavior rules for the cold-email skill pack
This file documents how the cold-email skill should behave when an agent (Claude / Cursor / Codex / any MCP-aware runtime) is operating inside a project that has installed this pack.
These rules are LOAD-BEARING. Skills check these on activation and adjust behavior accordingly.
Identity layering
Three files combine to define how the agent behaves:
| File | Job | Owner | Per project? |
skills/cold-email/SKILL.md (skill pack) | The JMC FRAMEWORK + structural rules | JMC (do not edit) | No — global |
SOUL.md (project root) | The OPERATOR'S VOICE | You | Yes |
brand-config.json (project root) | The OPERATOR'S BRAND CONFIG | You | Yes |
The skill enforces the framework. SOUL + brand-config personalize the output. Never let SOUL override the framework's banned-patterns list.
Rules of engagement
1. Always read brand-config.json + SOUL.md first
Before any output, agent must load both files (if they exist) into
context. If either is missing, run the setup mode (skills/cold-email/modes/setup.md) —
don't generate generic output.
2. Refuse to write generic output
If the operator has not set up brand-config.json + SOUL.md AND has not
explicitly opted out (e.g. --no-config), the agent refuses to draft
real outreach. Generic output is worse than no output. Scoring a draft the operator pasted is the exception: score it, say which checks the missing config skipped, then offer setup.
3. Quote signals verbatim
When using a public signal, quote the signal in the operator's source (LinkedIn post / job listing / press release). Do not paraphrase.
4. No fabrication
Never fabricate:
- Recipient names, roles, or companies
- Public signals that didn't happen
- Case-study outcomes
- Metrics or proof points
- DNS records (when running deliverability)
If a required input is missing, ask.
5. Self-check every output
Every output runs through the self-check rubric in skills/cold-email/SKILL.md
before delivery. If checks fail, regenerate before showing the user.
6. Surface drafts as drafts
Generated cold emails are drafts. Never send. Never auto-send. Never connect to send APIs without explicit per-call authorization.
7. Subagents stay in their lane
When the skill routes to a mode, that mode owns the workflow. Other modes do not interject mid-flow.
8. Respect rate limits
When running the deliverability mode against blacklist zones, respect their rate limits. Default to 1 lookup per second. Cache results.
9. Log decisions for audit
If an experiment log path is configured (operations.experiment_log_path
in brand-config), log every cold-email draft + every nurture sequence
- every audit run with timestamp + inputs + output summary. The operator can review.
10. Tell the user when you're guessing
If signal / pain / EVP / vocabulary are uncertain, flag the uncertain parts and ask for the actual data. Don't paint over uncertainty.
What the agent should NEVER do
- Write cold emails without a brand-config.json (refuse + route to onboarding)
- Use banned patterns from
skills/cold-email/references/banned-patterns.mdeven if the operator requests them — push back with the JMC alternative - Send any outreach without explicit per-call authorization
- Fabricate signals, recipients, or metrics
- Override the framework's structural constraints (≤90 words / binary CTA / etc.) even if SOUL.md asks for longer / softer
- Call paid APIs
Onboarding flow (first invocation)
If brand-config.json and SOUL.md are both missing on first
invocation:
- Welcome message: "I see you've installed claude-cold-email but haven't set up your brand-config or voice yet. Want me to walk you through it now? (~10 minutes)"
- If yes → run
/cold-email:cold-email setup(skills/cold-email/modes/setup.md) - If no → minimal-mode: agent will produce framework-shaped outputs but won't personalize them. Warn the operator that the output will be generic.
Telemetry / privacy
No telemetry. No outbound calls except:
- DNS lookups (the deliverability skill)
- Public blacklist APIs (the deliverability skill, when requested)
The pack does not call back to JMC Go-to-Market Systems servers.
Public sample and data policy (Jay, 2026-10-23)
Do not commit or distribute CSV files, proprietary datasets or bulk data exports in any format. Never repackage withdrawn rows as JSON, Markdown or XLSX. A CSV for Jay requires his explicit approval after inspecting its exact contents; do not create sample/export files automatically. Scorers may still read the operator's own private inputs. Tests use native fictional dictionaries and mocked file/parser contracts, without serializing a CSV sample or creating a CSV file.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

