Claude Code GTM repo layout

--- name: claude-code-gtm-repo-layout description: Scaffold a Claude Code / Codex / Cursor project folder for a GTM team (skills, agents, commands, hooks, rules, context folders). Use at the start of a new GTM workspace. source: Pattern described publicly by Dan Rosenthal (workflows.io) on X, https://x.com/dan__rosenthal/status/2058261757672538541 . Wording below is Agent Coach's, not his. --- # GTM workspace layout Create (or propose, if files already exist) this structure in the current repo, then stop and show the tree: - `CLAUDE.md` (or `AGENTS.md`): who we sell to (ICP), offer, tone of voice, banned claims, which tools/MCPs exist, approval rules. - `INDEX.md`: one line per skill/command so the agent can find them. - `.claude/skills/<skill>/SKILL.md`: one task per skill, end to end (e.g. `icp-research`, `list-build`, `enrich`, `first-line`, `sequence-draft`, `crm-writeback`). - `.claude/agents/`: longer multi-step workflows that a skill can hand off to. - `.claude/commands/`: slash commands that chain skills (e.g. `/campaign` = list-build → enrich → first-line → sequence-draft). - `.claude/hooks/`: approval gates. Anything that sends email, writes to the CRM, or spends enrichment credits must require a human "yes". - `.claude/rules/`: per-folder behavior (e.g. files under `clients/` are confidential). - `wiki/`, `clients/`, `raw-context/`, `archive/`: knowledge, per-client context, raw transcripts/exports, retired material. Rules: do not invent ICP facts; leave TODO placeholders where the owner must fill in. Do not create API keys or connect accounts.