sales-cadence-installer

Sales Cadence Installer

This skill walks a new user through a one-time setup, then generates four artifacts tailored to them:

1. Two scheduled tasks (sales-evening-recap and sales-morning-outreach) created via the scheduled-tasks MCP
2. hubspot-crm skill — installed at ~/.claude/skills/hubspot-crm/SKILL.md
3. A brand-voice skill — installed at ~/.claude/skills/<brand-skill-name>/SKILL.md
4. A personal-voice skill — installed at ~/.claude/skills/<voice-skill-name>/SKILL.md

Plus seed files at ~/.claude/sales-pipeline/:

• state.json (pipeline state)
• smtp.json (Gmail app password — placeholder; user fills in)
• hubspot.json (HubSpot access token — placeholder; user fills in)

Templates Live In This Skill's Folder

All templates are under templates/ next to this file. Read them with the Read tool when you need them. Use Bash with dirname and realpath if you need to resolve the absolute path to this skill folder, or just construct paths starting from $HOME/.claude/plugins/ if installed via plugin (the exact path depends on plugin install location).

Templates available:

• templates/evening-recap.md.tmpl
• templates/morning-outreach.md.tmpl
• templates/crm-hubspot.md.tmpl
• templates/brand-voice.md.tmpl
• templates/personal-voice.md.tmpl
• templates/state.json.tmpl

Reference docs (load only when relevant):

• references/hubspot-setup.md — how the user creates a HubSpot private app + token
• references/smtp-setup.md — how to generate a Gmail app password
• references/fallback-google.md — Google Tasks / Google Keep fallback when Apple Reminders / Granola are unavailable
• references/alternative-workflow.md — alternative workflows if the user doesn't want the standard cadence

Pre-Flight Checks (Step 0)

Before the interview, run these checks and tell the user what's missing:

which gws        # Google Workspace CLI — required for Gmail / Calendar / Drive

which remindctl  # optional — only needed on macOS for Apple Reminders

uname -s         # detect macOS vs Linux

If gws is missing, stop and tell the user: "The Google Workspace CLI (gws) isn't installed. The cadence reads Gmail, Calendar, and Drive through it. Install it first, then re-run setup. Want me to point you at install docs?"

If remindctl is missing on macOS, note it — you'll offer the Google Tasks fallback during the interview.

If on Linux/Windows, Apple Reminders isn't available — only offer Google Tasks or "skip reminders" options.

The Interview (Step 1)

Run this as a back-and-forth conversation. Ask one section at a time. Don't dump all questions at once. Save answers as you go in a working dict.

Privacy note: Never ask the user to paste API tokens or passwords in chat. For HubSpot tokens and Gmail app passwords, point them at the relevant reference docs and tell them where to write the file themselves.

Section A — Identity

Quick interview — about 10 minutes. I'll ask in chunks.

First, the basics:
1. Your first name (used in greetings and recap dialogue)?
2. Your full name (used in the personal-voice skill)?
3. Your work email?
4. Your phone (for sales email signatures — leave blank to skip)?

Capture: USER_FIRST_NAME, USER_FULL_NAME, USER_EMAIL, USER_PHONE.

Section B — Company

About your company:
5. Company name?
6. Primary domain (e.g. acme.com)?
7. One-line description of what your company does (e.g. "B2B AI services" or "marketing automation for dental practices")?

Capture: COMPANY_NAME, COMPANY_DOMAIN, COMPANY_DESCRIPTION.

Section C — Timezone & Schedule

8. Timezone? (e.g. "US Eastern", "US Pacific", "Europe/London")
9. What time should the morning outreach run? (default 8:30am)
10. What time should the evening recap run? (default 6:00pm)

Capture: TIMEZONE (display name), TIMEZONE_IANA (resolve to IANA tz id, e.g. America/New_York), MORNING_RUN_TIME, EVENING_RUN_TIME.

Section D — CRM (HubSpot)

The cadence syncs to HubSpot. You'll need a HubSpot Private App access token.

11. Do you have a HubSpot Private App created already, or should I walk you through it?

If they need help, read references/hubspot-setup.md and present the steps. Tell them to save the token to ~/.claude/sales-pipeline/hubspot.json in this format:

{"access_token": "their-token-here"}

Don't proceed past this point until they confirm they've created ~/.claude/sales-pipeline/hubspot.json.

python3 -c "import json,os,sys; p=os.path.expanduser('~/.claude/sales-pipeline/hubspot.json'); print('exists' if os.path.exists(p) and json.load(open(p)).get('access_token') else 'missing')"

If missing, tell them and wait.

Section E — Email & SMTP

The morning summary is sent to you via Gmail SMTP.

12. Confirm your Gmail address for sending the summary: {{USER_EMAIL}}?
13. Do you have a Gmail App Password set up?

If no, walk them through references/smtp-setup.md to create one. Tell them to save to ~/.claude/sales-pipeline/smtp.json as:

{"email": "{{USER_EMAIL}}", "gmail_app_password": "abcd efgh ijkl mnop"}

Verify exists:

python3 -c "import json,os; p=os.path.expanduser('~/.claude/sales-pipeline/smtp.json'); d=json.load(open(p)) if os.path.exists(p) else {}; print('ok' if d.get('email') and d.get('gmail_app_password') else 'missing')"

Section F — Reminders

For action items (e.g. "send proposal Friday"), I can use:

A) Apple Reminders — best on Mac, requires `remindctl` CLI
B) Google Tasks — cross-platform, uses your Google Workspace CLI
C) Skip reminders — track action items only in pipeline state

Which would you like? (A / B / C)

Based on choice, set the following variables:

Based on choice, set the following variables. (Newlines shown as \n mean an actual newline character — keep the formatting exact.)

Choice	REMINDER_TOOL	REMINDER_TOOL_PREFIX	REMINDER_LIST_NAME
A	remindctl	remindctl	Sales
B	gws tasks	gws	Sales
C	none	none	none

Choice	REMINDER_TOOL_BASH_CLAUSE	REMINDER_OPS_USE_LINE
A	, or remindctl``	- Use remindctl for all reminder operations.\n
B	`` (empty — gws already in list)	- Use gws tasks for all reminder operations.\n
C	`` (empty)	`` (empty)

Choice	REMINDER_CONFIG_LINE
A	- **Reminders**: Apple Reminders ("Sales" list) via remindctl``
B	- **Reminders**: Google Tasks ("Sales" list) via gws tasks``
C	- **Reminders**: disabled — action items tracked in state.json only

Render REMINDER_LIST_INIT_BLOCK, REMINDER_CHECK_BLOCK, and REMINDER_TOOL_DOCS accordingly:

Choice A (Apple Reminders):

• REMINDER_LIST_INIT_BLOCK:

  ```bash
  remindctl list "Sales" --create
  

  This is idempotent — if the list already exists, it's a no-op.

• REMINDER_CHECK_BLOCK:

  Check today's and overdue sales reminders:
  ```bash
  remindctl show today --list Sales
  

  remindctl show overdue --list Sales
  

• REMINDER_TOOL_DOCS: "the remindctl man page"

Choice B (Google Tasks):

• REMINDER_LIST_INIT_BLOCK:

  ```bash
  gws tasks tasklists list
  

  If a list named "Sales" doesn't exist, create it:

  gws tasks tasklists insert --params '{"requestBody": {"title": "Sales"}}'
  

  Idempotent — if the list already exists, the create call returns the existing list.

• REMINDER_CHECK_BLOCK:

  Check tasks due today or overdue:
  ```bash
  gws tasks tasks list --params '{"tasklist": "TASKLIST_ID", "showCompleted": false, "dueMax": "TODAY_END_ISO"}'
  

• REMINDER_TOOL_DOCS: "gws tasks reference"

Choice C (None):

• REMINDER_LIST_INIT_BLOCK: "Reminders disabled — track action items in state.json under each lead's pending_actions array. Skip the rest of this step."
• REMINDER_CHECK_BLOCK: "Reminders disabled — there are no reminders to check. Skip the rest of this step entirely (the 'For each incomplete/completed reminder' blocks below do not apply). Move directly to the next step. Use the pending_actions array in state.json as the source of truth for outstanding action items."
• REMINDER_TOOL_DOCS: "(disabled)"

Section G — Meeting Notes Source

For pulling meeting transcripts into the recap:

A) Granola (https://granola.ai) — primary if you use it
B) Google Drive only — find call notes by company name
C) Skip meeting notes

Which? (A / B / C)

Set MEETING_NOTES_BLOCK:

Choice A (Granola): Insert the Granola integration block (uses Granola MCP tools list_meetings, get_meetings).

Choice B (Google Drive only): Insert a block that searches Drive for files with the company name and reads the most recent ones.

Choice C (skip): Insert "Meeting notes disabled. Fall back to asking the user directly during recap."

(Full text for each block is in references/meeting-notes-blocks.md — read it when you need to substitute.)

Section H — Brand Voice

Tell the user this section drives the brand-voice skill. Run the questions in this order. Capture answers as multi-line strings.

14. Positioning — one sentence describing how you want to be perceived (e.g. "AI-as-a-service for SMBs", "the easiest accounting tool for freelancers"):
15. Primary market segment:
16. Secondary market segment (or "none"):
17. Products — name + one-line description for each, separated by newlines:
18. Taglines — list each tagline and where it gets used (hero, footer, etc.):
19. Three to five "we are" descriptors (e.g. "direct, warm, confident"):
20. Three to five "we are NOT" descriptors (e.g. "corporate, robotic, jargon-heavy"):
21. Cardinal rules — non-negotiable brand rules. List 3-7 (e.g. "Always lowercase 'i' in 'Ai'", "No claims that can't be proven"):
22. Banned words/phrases specific to your brand (e.g. industry buzzwords you refuse to use):
23. One on-brand example (a sentence/paragraph that nails the voice):
24. One off-brand example with explanation of why it fails:

Capture all answers. These become the {{...}} substitutions in brand-voice.md.tmpl.

Pick a brand skill name based on company name: <company-slug>-brand (e.g. acme-brand). Confirm with user.

Section I — Personal Voice

Now your personal sales voice. I'll learn it from real examples.

25. Paste 3 examples of YOUR sales emails — actual ones you've sent. Cold outreach, follow-ups, intro replies, anything that sounds like "you" at your most authentic. Don't sanitize — I want the real voice.

(Paste them now, separated by --- on its own line.)

Wait for the user to paste. Then analyze the examples and extract:

• Greeting pattern (e.g. "Hey [First Name],")
• Average word count → set TARGET_LENGTH (e.g. "60-80")
• Sentence structure observations
• Vocabulary they actually use (VOCABULARY_USED)
• Vocabulary they conspicuously avoid (VOCABULARY_AVOIDED) — derive from common AI/sales clichés they didn't use
• Sign-off block (SIGNOFF_BLOCK) — copy verbatim from one example
• 1-2 sentence voice snapshot (VOICE_SNAPSHOT)

Pick a voice skill name: talk-like-<first-name> (e.g. talk-like-jane). Confirm with user.

Based on those examples, here's what I picked up:
- Greeting: [observation]
- Length: ~[N] words per email
- Tone: [observation]
- Sign-off pattern: [block]

Look right? Anything to add or change?

Iterate until they're satisfied.

Section J — Internal Team Filter

26. Internal team email addresses to filter OUT of sales discovery (e.g. "[email protected], [email protected]"):
27. Custom Gmail discovery query (optional — e.g. "from:partner-domain.com" if you have a referral source):

Capture: INTERNAL_TEAM_EMAILS, CUSTOM_DISCOVERY_QUERY (or empty string).

Then derive the conditional rendering variables:

• INTERNAL_TEAM_FILTER_CLAUSE:
  • If INTERNAL_TEAM_EMAILS non-empty: (to/from <value>) ← note leading space
  • If empty: `` (empty string — the line still reads naturally)
• CUSTOM_DISCOVERY_QUERY_LINE:
  • If CUSTOM_DISCOVERY_QUERY non-empty: 3. <value> (one full numbered list line)
  • If empty: `` (empty string — the list ends at item 2)
• USER_PHONE_LINE:
  • If USER_PHONE non-empty: - **<USER_FIRST_NAME>'s phone**: <USER_PHONE>\n ← include the trailing newline
  • If empty: `` (empty string — line is dropped entirely)
• SECONDARY_MARKET_LINE:
  • If SECONDARY_MARKET non-empty and not literally none: - **Secondary market:** <value>
  • If empty or none: `` (empty)

Section K — Confirm and Build

Show a summary of all collected values and ask for final confirmation:

Ready to build. I'll:
- Create skill: hubspot-crm
- Create skill: <brand-skill-name>
- Create skill: <voice-skill-name>
- Create scheduled task: sales-evening-recap (runs daily at <evening-time> {{TIMEZONE}})
- Create scheduled task: sales-morning-outreach (runs daily at <morning-time> {{TIMEZONE}})
- Initialize: ~/.claude/sales-pipeline/state.json
- Verify: ~/.claude/sales-pipeline/smtp.json (already exists)
- Verify: ~/.claude/sales-pipeline/hubspot.json (already exists)

Proceed? (yes/no)

The Build (Step 2)

Once they confirm, do all of the following. Each is a separate operation; if one fails, report and continue with the rest where possible.

B1. Create directory tree

mkdir -p ~/.claude/sales-pipeline ~/.claude/skills/hubspot-crm

Also create dirs for the brand and voice skills:

mkdir -p ~/.claude/skills/<brand-skill-name> ~/.claude/skills/<voice-skill-name>

B2. Render and write the HubSpot skill

1. Read templates/crm-hubspot.md.tmpl
2. Substitute {{COMPANY_NAME}} and any other placeholders
3. Write to ~/.claude/skills/hubspot-crm/SKILL.md (use the Write tool)

B3. Render and write the brand-voice skill

1. Read templates/brand-voice.md.tmpl
2. Substitute every {{...}} placeholder with collected values
3. Write to ~/.claude/skills/<brand-skill-name>/SKILL.md

For multi-line answers (cardinal rules, banned words, etc.), preserve formatting as bulleted lists. Example for CARDINAL_RULES:

### 1. {{rule_1_title}}
{{rule_1_body}}

### 2. {{rule_2_title}}
{{rule_2_body}}

B4. Render and write the personal-voice skill

1. Read templates/personal-voice.md.tmpl
2. Substitute placeholders, including the three example emails
3. Write to ~/.claude/skills/<voice-skill-name>/SKILL.md

B5. Render and write state.json

1. Read templates/state.json.tmpl
2. Substitute identity placeholders
3. Write to ~/.claude/sales-pipeline/state.json ONLY IF it doesn't already exist (don't overwrite an existing one)

python3 -c "import os,sys; p=os.path.expanduser('~/.claude/sales-pipeline/state.json'); print('exists' if os.path.exists(p) else 'missing')"

If exists: ask the user "state.json already exists at ~/.claude/sales-pipeline/state.json — keep it or overwrite?" Default to keep.

B6. Render the two routine templates

1. Read templates/evening-recap.md.tmpl and templates/morning-outreach.md.tmpl
2. Substitute every placeholder including the conditional blocks (REMINDER_LIST_INIT_BLOCK, REMINDER_CHECK_BLOCK, MEETING_NOTES_BLOCK)
3. Hold the rendered text in memory — do NOT write to disk yet. Scheduled tasks live inside the scheduled-tasks MCP, not on disk as standalone files.

B7. Create the scheduled tasks

Use the scheduled-tasks MCP tool mcp__scheduled-tasks__create_scheduled_task for each:

Evening recap:

{
  "name": "sales-evening-recap",
  "schedule": "<cron derived from EVENING_RUN_TIME and TIMEZONE_IANA>",
  "prompt": "<full rendered evening-recap text>",
  "description": "Interactive evening sales recap"
}

Morning outreach:

{
  "name": "sales-morning-outreach",
  "schedule": "<cron derived from MORNING_RUN_TIME and TIMEZONE_IANA>",
  "prompt": "<full rendered morning-outreach text>",
  "description": "Autonomous morning sales outreach + summary email"
}

If the scheduled-tasks MCP isn't available in the user's environment, write the rendered routines to ~/.claude/scheduled-tasks/sales-evening-recap/SKILL.md and ~/.claude/scheduled-tasks/sales-morning-outreach/SKILL.md as a fallback, and tell the user how to register them with whatever scheduler they prefer.

B8. Confirm + smoke test

Print a confirmation summary:

Setup complete.

Skills installed:
- ~/.claude/skills/hubspot-crm/SKILL.md
- ~/.claude/skills/<brand-skill-name>/SKILL.md
- ~/.claude/skills/<voice-skill-name>/SKILL.md

Scheduled tasks created:
- sales-evening-recap (daily at <evening-time> <timezone>)
- sales-morning-outreach (daily at <morning-time> <timezone>)

Pipeline state:
- ~/.claude/sales-pipeline/state.json (initialized)
- ~/.claude/sales-pipeline/smtp.json (verified)
- ~/.claude/sales-pipeline/hubspot.json (verified)

Want me to run the evening recap once now in dry-run mode (no CRM writes, no reminder writes) so you can see it in action?

If yes, manually run through the evening recap logic — but for every Bash command that would write data (CRM, reminders, state.json), print "[DRY RUN] would run: " instead of executing it. Read-only operations (Gmail search, Calendar list, Drive list) run normally.

Substitution Reference

When rendering templates, every {{NAME}} token must be replaced. Here's the full list of variables and how to derive them:

Variable	Source	Example
USER_FIRST_NAME	Section A	Jane
USER_FIRST_NAME_UPPER	derived	JANE
USER_FULL_NAME	Section A	Jane Doe
USER_EMAIL	Section A	[email protected]
USER_PHONE	Section A	+1 555-555-5555 (or empty)
COMPANY_NAME	Section B	Acme Corp
COMPANY_DOMAIN	Section B	acme.com
COMPANY_DESCRIPTION	Section B	a B2B fintech
TIMEZONE	Section C	US Eastern
TIMEZONE_IANA	Section C resolved	America/New_York
MORNING_RUN_TIME	Section C	8:30am
EVENING_RUN_TIME	Section C	6:00pm
REMINDER_TOOL	Section F	remindctl / gws tasks / none
REMINDER_TOOL_PREFIX	Section F	remindctl / gws / none
REMINDER_LIST_NAME	Section F	Sales
REMINDER_TOOL_BASH_CLAUSE	Section F derived	, or \remindctl`` / empty / empty
REMINDER_OPS_USE_LINE	Section F derived	bullet line + newline / empty
REMINDER_LIST_INIT_BLOCK	Section F derived	(block above)
REMINDER_CHECK_BLOCK	Section F derived	(block above)
REMINDER_TOOL_DOCS	Section F	"the remindctl man page"
REMINDER_CONFIG_LINE	Section F derived	full bullet line, see table above
INTERNAL_TEAM_FILTER_CLAUSE	Section J derived	(to/from <emails>) or empty
CUSTOM_DISCOVERY_QUERY_LINE	Section J derived	3. <query> or empty
USER_PHONE_LINE	Section A derived	- **<First>'s phone**: <num>\n or empty
SECONDARY_MARKET_LINE	Section H derived	- **Secondary market:** <val> or empty
MEETING_NOTES_BLOCK	Section G	(block from references/meeting-notes-blocks.md)
CRM_NAME	constant for HubSpot	hubspot-crm
CRM_BASH_PREFIX	constant for HubSpot	curl
BRAND_SKILL_NAME	Section H	acme-brand
VOICE_SKILL_NAME	Section I	talk-like-jane
POSITIONING	Section H	(user input)
PRIMARY_MARKET	Section H	(user input)
SECONDARY_MARKET	Section H	(user input or "none")
PRODUCTS_LIST	Section H	(formatted bulleted list)
TAGLINES_TABLE	Section H	(formatted markdown table)
POSITIVE_DESCRIPTORS	Section H	(user input)
NEGATIVE_DESCRIPTORS	Section H	(user input)
CARDINAL_RULES	Section H	(formatted numbered list)
BANNED_WORDS	Section H	(formatted bulleted list)
VOICE_TONE_RULES	Section H derived	(synthesize from W/W-NOT)
ON_BRAND_EXAMPLES	Section H	(user input, formatted)
OFF_BRAND_EXAMPLES	Section H	(user input, formatted)
VOICE_SNAPSHOT	Section I derived	(1-2 sentence summary of their style)
GREETING_PATTERN	Section I derived	(extracted from examples)
SENTENCE_STRUCTURE	Section I derived	(extracted from examples)
VOCABULARY_USED	Section I derived	(extracted from examples)
VOCABULARY_AVOIDED	Section I derived	(inferred — common AI/sales words not present)
SIGNOFF_BLOCK	Section I derived	(extracted verbatim)
TARGET_LENGTH	Section I derived	60-80
EXAMPLE_1 / _2 / _3	Section I	(user paste)
INTERNAL_TEAM_EMAILS	Section J	[email protected], [email protected]
CUSTOM_DISCOVERY_QUERY	Section J	(user input or empty)

Error Recovery

• User pastes a token in chat by mistake: Tell them to revoke it, generate a new one, and write it to file directly. Do NOT save the chat-pasted token to disk.
• Scheduled-task MCP fails: Fall back to writing the routines to ~/.claude/scheduled-tasks/<name>/SKILL.md and tell the user.
• Template file not found: Check the install location. The plugin may be installed under ~/.claude/plugins/sales-cadence/skills/sales-cadence-installer/templates/. Use ls to find the actual path.
• User wants to re-run setup: Re-running this skill is safe — it doesn't overwrite state.json if it exists, and it asks before overwriting any other file.

When NOT To Use This Skill

• The user already has the cadence running and just wants to TWEAK something — direct them to edit the relevant skill or scheduled task in place. Re-running setup is overkill.
• The user wants a different CRM than HubSpot — this skill only ships the HubSpot adapter. Tell them and offer to extend it after install.
• The user wants a totally different sales workflow — read references/alternative-workflow.md and discuss with them what they actually want, then decide whether this skill applies or whether to build something custom.