gtm-pipeline
AGENTS.md16 starsChanged 49 days agoApache-2.0View on GitHub
This file is tool-neutral: it applies to Claude Code, Codex, Cursor, or any agent that can read files and run shell commands. CLAUDE.md adds the Claude Code specifics. REVIEW.md is the quality bar every run clears; read it before a live run. A plain-English campaign brief becomes a deduped, qualified, enriched, sequencer-ready contact list. One list_id threads through these stages: - Entry point: follow agents/orchestrator/agent.md. Each stage is a markdown procedure in agents/<stage>/agent.md. They are plain instructions, so any agent…
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 33,315 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 repository10 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- activateSkill
- company-discoverySkill
- company-enricherSkill
- contact-qualifierSkill
- contact-sourcerSkill
- email-finderSkill
- gtm-setupSkill
- gtmSkill
- phone-finderSkill
The file
We split this AGENTS.md into the rules it gives an agent
7 rules. About 0 are common, 2 are distinctive, and 5 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
Distinctive · estimate
This author's own choices. Worth a look.
5
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 19
Entry point: follow agents/orchestrator/agent.md.
- Very specific: line 22
Storage: every read and write goes through python3 storage/cli.py <op> --input '<json>'.
- Very specific: line 33
Read the verdict, not the exit code: suppress (research posture) exits 0 with applied: false when it could not check anything.
- Very specific: line 35
After each stage, compare list_summary with what the stage reported, and log_event it.
- Very specific: line 36
Never push to a sequencer unless preflight_activate returns ok: true (exit 0).
- Distinctive: line 37
Never invent an email, phone, title, URL or company fact.
- Distinctive: line 38
Never edit an agent prompt to support a provider.
≈ “Common” is an estimate: at least 1 in 5 of the 42,283 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 — the contract for any coding agent
This file is tool-neutral: it applies to Claude Code, Codex, Cursor, or any agent that can
read files and run shell commands. CLAUDE.md adds the Claude Code specifics.
REVIEW.md is the quality bar every run clears; read it before a live run.
What this repo does
A plain-English campaign brief becomes a deduped, qualified, enriched, sequencer-ready contact
list. One list_id threads through these stages:
company_search → company_enrich → people_search → [suppress] → qualify → [qa] →
email_enrich → phone_enrich → [qa · suppress(send) · preflight] → activateHow to run it
- Entry point: follow
agents/orchestrator/agent.md. Each stage is a markdown procedure inagents/<stage>/agent.md. They are plain instructions, so any agent can follow them. - Storage: every read and write goes through
python3 storage/cli.py <op> --input '<json>'. It reads the backend fromgtm.config.yaml. Never write raw SQL or edit files under.gtm-data/by hand. Ops, inputs and exit codes are listed at the top ofstorage/cli.py. - Fan-out: Claude Code runs the parallel stages as workflows in
.claude/workflows/. An agent without a workflow runtime runs the same prompt once per item, sequentially, and must still report the items that failed. - Providers: chosen by
gtm.config.yamlwaterfalls and described byproviders/<name>/manifest.yaml. Keys come from the local environment only.
Rules that are easy to break
- Read the verdict, not the exit code:
suppress(research posture) exits 0 withapplied: falsewhen it could not check anything. - After each stage, compare
list_summarywith what the stage reported, andlog_eventit. - Never push to a sequencer unless
preflight_activatereturnsok: true(exit 0). - Never invent an email, phone, title, URL or company fact. Blank is a valid answer.
- Never edit an agent prompt to support a provider; edit the manifest.
Checks before you commit
bash scripts/selftest.sh # storage, gates, adapters, workflow syntax, sweep tests
bash scripts/scrub-check.sh # no secrets, no private stringsDoes 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.

