Sales Outreach

Cold Outreach Sequence Generator

Tier 0 Doctrine

• Tier: Spoke (subagent of sales-prospect as sales-strategy; also standalone via /sales outreach <prospect>)
• Authority: Reads discovery briefing from orchestrator or performs own research; scores Outreach Readiness dimension only when subagent
• Determinism first: Framework selection uses a declared decision tree. Scoring uses declared formulas and calibration tables. No LLM interpretation of scores.
• No fabrication: Every personalization element must reference specific, verifiable prospect data. Absence of data is scored, not invented.

IMO (Top-Level)

Constants

Variables

Workflow

BLOCK 1: Personalization Research

Governed by: C&V

Constants: 3 trigger categories (Company 7 types, Personal 6 types, Industry 5 types), trigger quality scale (Hot/Warm/Cool), trigger freshness windows, banned openers, first line rule Variables: company_triggers, personal_triggers, industry_triggers, trigger_quality_ratings, best_opening_angle, secondary_angle

IMO:

• Ingress: Prospect identifier validated as non-empty string. If discovery briefing present, validate as parseable markdown.
• Middle:
  • If discovery briefing contains pre-fetched data, skip redundant research for covered dimensions
  • Execute company trigger searches: "[company]" funding, "[company]" product launch, "[company]" hiring, "[company]" partnership, "[company]" expansion, "[company]" news, "[company]" CEO interview
  • Execute personal trigger searches: "[contact]" "[company]" LinkedIn, "[contact]" "[company]" interview, "[contact]" "[company]" presentation, "[contact]" "[company]" article
  • Execute industry trigger searches: "[industry]" trends, "[industry]" challenges, "[industry]" regulation, "[industry]" disruption, "[industry]" market shift
  • For each trigger found: record description, date, category, quality rating (Hot/Warm/Cool per freshness windows)
  • Select best opening angle: highest-quality trigger with strongest connection to prospect
  • Select secondary angle: next-best trigger from a different category
  • Compile personalization research summary: company triggers, personal triggers, industry triggers, best/secondary angles
• Egress: Personalization research summary with all triggers rated and angles selected

Go/No-Go: Proceed if at least one trigger found at any quality level. If zero triggers found across all 3 categories, note limitation and proceed with industry-level personalization (weakest approach). Never halt — always generate sequence.

BLOCK 2: Framework Selection & Sequence Generation

Governed by: IMO

Constants: 4 outreach frameworks with structure and selection criteria, framework decision tree, 5-email sequence structure (names, days, goals, word limits), email writing rules (subject limits, body limits, CTA rules, tone rules, banned jargon, banned openers, first line rule), A/B test rule Variables: selected_framework, email_subjects_ab, email_bodies, case_study_match

IMO:

• Ingress: Personalization research summary from Block 1
• Middle:
  • Evaluate decision tree top-to-bottom:
    1. Hot company trigger present? -> Framework 3 (Trigger Event): reference trigger + create urgency
    2. Mutual connection available? -> Framework 4 (Mutual Connection): reference shared element + peer framing
    3. Strong personalization anchor (post, article, product)? -> Framework 1 (Observation-Connection-Ask): observation + bridge + soft ask
    4. Relevant case study with metrics? -> Framework 2 (Problem-Proof-Ask): problem + proof + ask
    5. None of above? -> Framework 1 with industry-level observation
  • Generate 5-email sequence following selected framework:
    • Email 1 — Hook (Day 1): specific observation first line, bridge to relevance, value statement with proof, soft CTA. Goal: get response.
    • Email 2 — Value Add (Day 3): brief reference to email 1 (no guilt), valuable insight/resource/data, why it matters to them, soft close. Goal: give, not ask.
    • Email 3 — Social Proof (Day 7): specific result for similar company, brief context, how achieved, bridge to prospect, slightly more direct CTA. Goal: trust via proof.
    • Email 4 — Different Angle (Day 14): new observation or pain point, different stakeholder perspective, fresh proof point, direct but respectful CTA. Goal: new perspective.
    • Email 5 — Breakup (Day 21): acknowledge busy (no guilt), restate core value in one sentence, leave door open. Under 75 words. Goal: respectful close.
  • For each email: generate 2 subject line variants (A/B), 2 opening line variants (A/B)
  • Enforce all writing rules: word limits, CTA count, tone, jargon ban, opener ban, first line specificity
  • If no matching case study for Email 3, substitute industry benchmark or third-party data
• Egress: Complete 5-email sequence with selected framework, A/B variants, per-email structure

Go/No-Go: All 5 emails must be generated regardless of personalization quality. Flag emails that rely on weak personalization. If running as subagent with time constraint, prioritize Email 1 quality over completeness of emails 4-5.

BLOCK 3: LinkedIn Integration & Channel Strategy

Governed by: Circle

Constants: 4 LinkedIn touchpoint slots (Day 0, 5, 10, 18), connection note limit (300 chars), DM limit (150 words), send timing lookup table, deliverability rules, follow-up rules by scenario Variables: linkedin_touchpoint_content, linkedin_profile_url, send_timing_recommendation, audience_type

IMO:

• Ingress: 5-email sequence from Block 2 + contact data from Block 1
• Middle:
  • Day 0 — Connection request: Write personalized note under 300 characters using best trigger. Never use default "I'd like to add you to my network."
  • Day 5 — Content engagement: Identify most recent LinkedIn post. Plan genuine, substantive comment (not "Great post!"). Add perspective or data. No pitch.
  • Day 10 — LinkedIn DM: Write cross-channel message under 150 words. Reference emails sent. Different value prop than emails. End with question.
  • Day 18 — Share content: Identify article/report/insight relevant to prospect. Plan post connecting content to their challenge. Tag only if genuinely relevant.
  • Timing: Classify contact into audience type (C-suite, Directors, Technical/IC, Founders). Look up send timing from constant table. Recommend specific days and times in prospect timezone.
  • Deliverability: Apply deliverability rules — recommend volume ramp, warmup period, domain setup, authentication checklist.
  • Follow-up logic: Map each response scenario (no response, OOO, not interested, not now, positive, question) to prescribed action from follow-up rules constant.
• Egress: LinkedIn touchpoint plan (4 entries with content) + send timing recommendation + deliverability checklist + follow-up decision table

Go/No-Go: Proceed unconditionally. If LinkedIn profile not found, omit touchpoints and note channel limitation in scoring.

BLOCK 4: Scoring & Output Assembly

Governed by: CTB

Constants: 4 scoring dimensions (0-25 each), score formula (sum = 0-100), calibration scale, output file name (OUTREACH-SEQUENCE.md), output schema Variables: personalization_depth_score, trigger_events_score, channel_strategy_score, message_market_fit_score, outreach_readiness_score

IMO:

• Ingress: All outputs from Blocks 1-3
• Middle:
  • Score each dimension 0-25 per calibration scale:
    • Personalization Depth: 20-25 strong personal + company triggers; 15-19 moderate; 10-14 weak; 0-9 none
    • Trigger Events Found: 20-25 hot trigger within 30 days; 15-19 warm within 90 days; 10-14 cool 3-6 months; 0-9 none
    • Channel Strategy Clarity: 20-25 LinkedIn + email + engagement path clear; 15-19 most channels; 10-14 email only; 0-9 no clear path
    • Message-Market Fit: 20-25 clear value prop + strong case study; 15-19 good match + moderate proof; 10-14 some relevance; 0-9 weak fit
  • Compute outreach_readiness_score = sum of 4 dimensions
  • Assemble OUTREACH-SEQUENCE.md with sections: Prospect Summary table, Personalization Research (company/personal/industry triggers), Selected Framework with reasoning, 5-email sequence (each with subject A/B, body, CTA, LinkedIn touchpoint, A/B variations), LinkedIn Touchpoint Summary table, Sending Best Practices, Objection Preparation table (3 likely objections with responses)
  • If standalone: write OUTREACH-SEQUENCE.md to working directory
  • If subagent: return structured block with score + full sequence for orchestrator
  • Generate terminal summary with Unicode bar charts for each dimension
• Egress: OUTREACH-SEQUENCE.md written (standalone) or structured score block returned (subagent)

Go/No-Go: Score must be calculable (all 4 dimensions scored). All 5 emails must be present. Any dimension without data: assign score per calibration, note data gap explicitly.

Rules

1. Never write any email before completing personalization research. Generic outreach is worse than no outreach.
2. Never start an email with "I hope this finds you well", "My name is...", or "I'm reaching out because..." — first line must reference specific prospect data.
3. Never include more than one CTA per email. Low friction, question format, specific.
4. Never use banned jargon (synergy, leverage, circle back, touch base, low-hanging fruit, deep dive, bandwidth, ecosystem).
5. Never guilt-trip in follow-ups or breakup emails. No "I've sent 4 emails" or "This is my last email."
6. Never exceed word limits: emails 1-4 under 100 words, email 5 under 75 words, LinkedIn DM under 150 words.
7. Never fabricate triggers, case studies, or personalization data. Missing data is scored and noted, not invented.
8. Never use HTML formatting, emojis, exclamation marks, or bold/italic in generated emails. Plain text only.
9. Never skip A/B variant generation. Every email requires 2 subject line variants and 2 opening line variants.
10. Never send the LinkedIn default connection request. Every connection note must be personalized under 300 characters.

Reference Pointers