team-init

name	team-init
description	Set up a new ESCC workspace — detect the connected GTM stack, write a tailored workspace CLAUDE.md. Trigger: 'init my workspace', 'set up ESCC for my team', /team-init. Installs = configure-escc.
origin	ESCC

Team Init

Sets up a workspace-local CLAUDE.md that captures the detected GTM stack, sender identity, ICP pointer, and persona. It is the first step any rep or team should run in a new project directory — before prospecting, before outreach, before anything else.

Safety posture: This skill does NOTHING to the CRM, sends nothing, and reads only locally available MCP tool-name metadata (not prospect data). Writing CLAUDE.md is a local-file action only. crm-operator remains the sole CRM writer.

When to Activate

• User says /team-init, "init my workspace", "set up my GTM stack", "configure my ESCC workspace".
• A new project directory lacks a CLAUDE.md and the user is beginning sales work.
• A team lead is onboarding reps and wants a standard baseline config written.
• The connected GTM stack has changed (new CRM, new email tool) and the workspace config is stale.

Do not activate for:

• General "how does X work" questions -- route to escc-guide.
• Installing ESCC components or changing install profiles -- route to configure-escc.
• Any action that reads or writes HubSpot -- that is crm-operator territory.

Workflow

Step 1 -- Detect the connected GTM stack

Inspect the names of currently available MCP tools (what tools are wired up in this session). This is a metadata-only read -- no prospect data is touched.

For each entry in config/gtm-stack-mappings.json, test whether any available tool name matches the indicator glob:

Indicator pattern	Stack
mcp__hubspot__*	HubSpot CRM (system of record)
mcp__*[Gg]mail*	Gmail (draft-only by construction)
mcp__*Google_Calendar*	Google Calendar
mcp__*[Ff]ireflies*	Fireflies (call transcripts)
mcp__*[Ss]lack*	Slack (alert delivery)
mcp__*[Ee]xa*	Exa (web research)
mcp__*firecrawl*	Firecrawl (web crawl/extract)
mcp__*[Aa]pollo*	Apollo (prospecting data)
mcp__*[Cc]lay*	Clay (enrichment)
mcp__*[Zz]apier*	Zapier (multi-app automation)

Collect every matched entry. Each match contributes its recommends block: skills, rules, hooks, and optionally a profile hint.

If NO tools are detected, record the workspace as "offline / unconnected" and note in the output that the user should wire up MCP tools and re-run.

Step 2 -- Collect sender identity and persona

Ask the user (or infer from git config / existing context) for the two items that personalize every skill downstream:

1. Sender identity (name, title, company, email address) -- this is the VOICE PROFILE header that outreach skills use. Example: Alex Kim, SDR, Example Co -- [e-mail]

2. Persona -- select one:

   • sdr -- prospecting, cold outreach, inbound triage, meeting booking
   • ae -- call prep, discovery, deal review, proposals, negotiation
   • sales-manager -- pipeline hygiene, forecast rollup, coaching
   • revops -- GTM stack ops, CRM hygiene, routing rules, reporting
   • full -- full ESCC surface (team leads, power users, solo founders)

If the user has already answered these in a prior session and .claude/escc/identity.json exists, read it and propose "keep existing / update" rather than re-asking.

Step 3 -- Resolve recommended surface from mappings

Merge the recommends blocks from all matched mappings into three lists:

• Skills to activate (deduplicated, sorted): the union of all matched skills arrays from config/gtm-stack-mappings.json.
• Rules to load (deduplicated): the union of all matched rules arrays.
• Hooks to enable (deduplicated): the union of all matched hooks arrays.
• Suggested base profile: if any mapping recommends a profile, surface the highest-privilege one (precedence: full > revops > ae > sdr), but let the user override with their chosen persona from Step 2.

Step 4 -- Dry-run summary

Before writing any file, output a clear plan showing:

Detected GTM stack:
  - HubSpot CRM (system of record) [mcp__hubspot__*]
  - Gmail draft-only [mcp__claude_ai_Gmail__*]
  - Fireflies transcripts [mcp__claude_ai_Fireflies__*]

Sender identity: Alex Kim, SDR, Example Co <[e-mail]>
Persona: sdr

Recommended skills: cold-outreach, email-outbound-ops, follow-up-ops,
  inbox-triage, outbound-sequences, reply-handling, account-research,
  account-memory, meeting-booking, discovery-notes, call-review
Recommended rules: outbound-compliance, messaging-style, data-handling,
  meeting-standards
Recommended hooks: pre:outbound-send-gate, post:outbound-style-check,
  post:crm-log-reminder

CLAUDE.md will be written to: ./CLAUDE.md
  (append-safe: existing file will be shown as diff before applying)

Proceed? [yes / edit / cancel]

Wait for confirmation. Do not write anything until the user says yes or approves.

Step 5 -- Write the workspace CLAUDE.md

On approval, write (or append-merge into) ./CLAUDE.md with this structure:

# Workspace: ESCC Sales Config
<!-- Generated by team-init -- re-run /team-init to refresh -->

## GTM Stack
<!-- detected from available MCP tool names -->
- HubSpot CRM (system of record): mcp__hubspot__*
- Gmail (draft-only): mcp__claude_ai_Gmail__*
- Fireflies (transcripts): mcp__claude_ai_Fireflies__*

## Sender Identity
Name: Alex Kim
Title: SDR
Company: Example Co
Email: [e-mail]

## Persona
sdr

## ICP Pointer
<!-- Edit this line to reference your ICP profile skill or context file -->
See: skills/icp-profile/SKILL.md

## Recommended Skills (auto-loaded)
cold-outreach, email-outbound-ops, follow-up-ops, inbox-triage,
outbound-sequences, reply-handling, account-research, account-memory,
meeting-booking, discovery-notes, call-review, meeting-followthrough

## Recommended Rules
outbound-compliance, messaging-style, data-handling, meeting-standards

## Recommended Hooks
pre:outbound-send-gate, post:outbound-style-check, post:crm-log-reminder

## ESCC Notes
- Gmail is draft-only by construction. No agent sends email.
- pre:outbound-send-gate fails CLOSED. Clean outbound-reviewer output
  is required before any live send.
- crm-operator is the only HubSpot writer.

If a CLAUDE.md already exists:

• Read it.
• Show the proposed diff (added ESCC section only).
• Do not overwrite any existing non-ESCC content.
• Append the ESCC section after existing content.

Step 6 -- Confirm and exit

After writing, output:

Workspace initialized.
  File written: ./CLAUDE.md
  Persona: sdr
  Stack: HubSpot + Gmail + Fireflies (3 tools)

Next steps:
  1. Run /prospect or escc:prospecting-pipeline to build your first account list.
  2. Run /sequence or escc:outbound-sequences to start an outbound play.
  3. Run /team-init again after connecting new MCP tools.
  4. To install additional ESCC components: escc:configure-escc

Examples

Clean first run:

/team-init
-> Detects: HubSpot + Gmail + Google Calendar
-> Asks: sender identity and persona (ae selected)
-> Dry-run summary shown, user approves
-> Writes CLAUDE.md with ae surface + HubSpot + email + calendar skills

Re-run after adding Fireflies:

/team-init
-> Detects: HubSpot + Gmail + Google Calendar + Fireflies (new)
-> Existing CLAUDE.md found -- shows diff
-> Adds: discovery-notes, call-review, meeting-followthrough
-> Appends diff only, existing CLAUDE.md content preserved

Offline / no MCP tools connected:

/team-init
-> No tool matches found
-> Output: "No GTM tools detected. Connect HubSpot, Gmail, or other
   MCP tools and re-run /team-init to get a tailored workspace config."
-> CLAUDE.md not written (nothing useful to write)

Anti-patterns

• Reading prospect data during detection. Stack detection reads TOOL NAMES only (metadata). Never call any MCP tool to detect the stack -- just inspect what tools are available by name.
• Writing CRM records. team-init writes ONE local file (CLAUDE.md). Any CRM mutation belongs to crm-operator, not here.
• Overwriting existing CLAUDE.md without diff review. Always show the proposed change and wait for approval before modifying an existing file.
• Conflating team-init with configure-escc. team-init writes a workspace config. configure-escc installs ESCC components (skills, rules, hooks). They are separate.
• Skipping the dry-run step. No file is written until the user approves the plan.

Related

• config/gtm-stack-mappings.json -- the indicator-to-recommendations map this skill reads.
• schemas/gtm-stack-mappings.schema.json -- the schema that validates the map.
• configure-escc -- installs ESCC components (skills/rules/hooks); complementary to team-init.
• escc-guide -- answers navigation and "how do I" questions.
• icp-profile -- the ICP definition the workspace config points to.
• Command: /team-init (thin shim delegating here).