agentleFS
andrerodriguesopsandrerodriguesops/gtm-engine/CLAUDE.md

gtm-engine

CLAUDE.md0 starsChanged 9 days agoMITView on GitHub

This repo is COMPANY's go-to-market system: the knowledge layer (skills) and the agent layer that runs on it. If the opening line above still holds the placeholder name, or SETUP-PROGRESS.md has an open [ ] box in steps 0 to 3a (a deferred [~] box does not count), the engine is not set up yet. At the start of the session, before anything else, tell the user that in one sentence and offer to walk them through setup. If they…

The file

We split this CLAUDE.md into the rules it gives an agent

8 rules. About 0 are common, 3 are distinctive, and 5 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    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
  1. Very specific:

    If the opening line above still holds the placeholder name, or SETUP-PROGRESS.md has an open [ ] box in steps 0 to 3a (a deferred [~] box …

    line 11
  2. Distinctive:

    If they ask for something else first, help with that, and mention that answers will be generic until setup step 1 is done.

    line 14
  3. Very specific:

    archive/ is not part of the dataset: Search it when hunting a past decision.

    line 56
  4. Very specific:

    Before writing any document, answer: *who reads this, and are we asking them for something?* docs/design-system.md routes the answer.

    line 62
  5. Distinctive:

    Read the relevant skill before producing client-facing output.

    line 64
  6. Distinctive:

    When a skill turns out to be wrong, fix the skill in the same session and commit with a message that says what changed and why.

    line 65
  7. Very specific:

    When something is unresolved, say so and add it to NEEDS.md.

    line 67
  8. Very specific:

    Run python3 ops/check-corpus.py before merging and python3 ops/check-client-copy.py before anything goes to a client.

    line 68

≈ “Common” is an estimate: at least 1 in 5 of the 27,641 CLAUDE.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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
CLAUDE.md

GTM system — agent instructions

<!-- Replace the placeholder on the line below with your company name, on that line only. Keep this file short: it loads into every session. -->

This repo is COMPANY's go-to-market system: the knowledge layer (skills) and the agent layer that runs on it.

First run

If the opening line above still holds the placeholder name, or SETUP-PROGRESS.md has an open [ ] box in steps 0 to 3a (a deferred [~] box does not count), the engine is not set up yet. At the start of the session, before anything else, tell the user that in one sentence and offer to walk them through setup. If they agree, run the setup skill. If they ask for something else first, help with that, and mention that answers will be generic until setup step 1 is done.

The architecture

Knowledge layer — .claude/skills/ and docs/. What is true and what to do. Read whole. Agent layer — agents/. One runbook per scheduled agent. Agents read the skills; they never embed knowledge of their own.

One source of truth per fact. A fact lives in one file; everything else points at it. The same fact written in two places is a defect even when the copies agree, because they will stop agreeing. Doctrine: docs/system-hygiene.md.

Two authorities, two questions.

QuestionAuthority
What we sell, and what it costs.claude/skills/sales-playbook/references/commercial.md
What exists, and what state it is in.claude/skills/product-context/SKILL.md

On a number, commercial.md wins.

Two corpora. .claude/skills/ and docs/ hold company knowledge: stable, corrected in place. clients/ holds client evidence: one folder per account, a permanent record, annotated rather than corrected. A client record cites a price; it never states one.

Non-negotiables

Source: docs/non-negotiables.md. The list below is a derived copy, kept here because this file loads automatically. Change a rule there first, then here.

  1. Never quote a price without reading commercial.md. A new or changed figure lands in commercial.md first.
  2. Never state a capability the product reference does not support, and never describe a planned feature as live.
  3. Never edit anything already in a client's hands unless the change cannot alter what they believe was offered. A factual change is a reissue.
  4. Never send email, contracts or client messages on a person's behalf. Agents write drafts.
  5. Never move a deal stage unattended. Agents propose; a person moves it. The one exception is a completed DocuSign envelope, which moves the deal to Closed Won (close/CLOSE.md).
  6. Never guess a fact about a prospect. It comes from enrichment or their own site, or it is cut.
  7. archive/ is not part of the dataset. Search it when hunting a past decision; never load it as truth.
<!-- Add your own: banned words, names you never say, legal lines. Each gets a ruling and a date. -->

Working style

  • Before writing any document, answer: who reads this, and are we asking them for something? docs/design-system.md routes the answer.
  • Read the relevant skill before producing client-facing output. Memory of this domain drifts.
  • When a skill turns out to be wrong, fix the skill in the same session and commit with a message that says what changed and why. Do not work around it.
  • When something is unresolved, say so and add it to NEEDS.md. Do not pick a plausible answer.
  • Run python3 ops/check-corpus.py before merging and python3 ops/check-client-copy.py before anything goes to a client.

Git

  • Skills are versioned like code. A pricing change must be findable from its commit message.
  • Tag whenever something goes to a client: client/<name>-<YYYY-MM-DD>.
  • Several sessions may run at once. Never commit, rebase or switch branches in a shared checkout: git worktree add, work there, push, remove. Exception: setup, run alone in a fresh clone, commits in place.
  • Scheduled agents push only to claude/<agent> branches, never to main.

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.