Cold Email Skill

Step-by-Step Workflow

Step 1: Load Context

Load f13i docs using load-f13i-docs skill, or read directly from the Obsidian vault with obsidian_get_note (format: content):

• 10-brain/05-writing.md (voice rules)
• 10-brain/02-business.md (pipeline, ICP, existing relationships)
• Voice docs: 10-brain/voice/01-voice-doc.md, 10-brain/voice/02-llm-prompt.md (fetch these for all outreach tasks)

Step 2: Warm Signal Check (ALWAYS run before drafting)

Check for existing relationship context in this order:

Granola:

Granola:query_granola_meetings
query: "[person name] OR [company name]"

Google Calendar:

Google Calendar:list_events
q: "[person name] OR [company name]"
timeMin: [3 months ago]
timeMax: [today]

Gmail:

Gmail:search_threads
query: "from:[company domain] OR to:[company domain]"

Obsidian vault (second brain):

obsidian_search_notes
query: "[company name]"

If ANY warm signal found → email MUST open with that specific moment/interaction, not a general insight. Override the hook decision tree.

Step 3: Research the Prospect

• Web search: "[person name] [company]" + "[company] pricing page"
• Fetch their pricing page directly if available
• If from a job posting: extract the hiring pain from the JD — that IS the hook

Step 4: Hook Selection (if no warm signal)

Pick ONE hook based on what the research surfaced:

Step 5: Draft the Email

Structure:

Subject: [3-6 words, specific to their situation — no generics]

[Warm signal anchor OR specific observation — 1 sentence]

[The so-what / what it means for them — 1-2 sentences]

[Credibility anchor — 1 sentence, specific to why you — tie to Grafana or relevant experience]

[No-strings-attached ask — 1 sentence]

[Signature]
Everyone's got a pricing opinion. Almost nobody has pricing data.
f13i.com | cal.com/farhan-m | me@f13i.com

Voice rules (from 10-brain/05-writing.md):

• Contractions always
• No hedging phrases ("I think", "might", "perhaps")
• No consulting jargon
• Short sentences, varied length
• Sound like a practitioner, not a vendor

Step 6: Grade the Email

Grade as the recipient would grade it (use their background/role):

Do not send below B. If B or below, fix it and re-grade before presenting final version.

Step 7: Email Verification

Before finalizing, note email format:

• Research the company's email pattern (first.last, firstlast, etc.)
• Recommend verifying via Hunter.io or NeverBounce free tier before sending
• Note the email domain's mail provider if findable (MX record lookup via web search)

ICP Reminder

Strongest current lane: companies embedding AI into their product who don't know how to price it. Secondary: Series A-B SaaS with a clear pricing problem visible from the outside. Avoid: AI-native operators running full Claude/GPT/Gemini stacks with full business context — they don't see consultant value.

JD-to-Client Playbook

When Farhan shares a job posting for a pricing role:

1. Extract the pain from the JD — what are they trying to solve?
2. Identify the hiring manager (usually the poster)
3. Research their current pricing page
4. Apply this workflow — JD pain becomes Hook B (left on table) or Hook A (bug)
5. Grade using the A+/B rubric before finalizing

Preflight — resolve tool signatures at runtime

Do not trust the tool names or parameter names written above. They drift.

Before the first Gmail, Calendar, Granola, or Obsidian call, run tool_search for each ("gmail search threads", "calendar list events", "granola meetings", "obsidian get note") and use the names and parameters it returns. The names here are correct as of 2026-07-30; the parameter schemas are not documented here on purpose, because hardcoding them is what broke this skill.

History: this skill shipped for months calling gmail_search_messages, gmail_read_thread, and gcal_list_events — none of which exist. It failed silently, producing confident output with the email and calendar legs missing entirely. Nothing reported it. Fixed 2026-07-30.

Failure behaviour: if a source fails, name it in the output. A partial result that says which leg died beats a clean-looking one that quietly dropped a source.