SKILL.md

---
name: offer-definer
description: >
  Extract every value proposition from a company's materials into a tiered, persona-mapped
  inventory, then frame the strongest ones into ready-to-use offers at three granularities
  (one-liner, value proposition, full offer). Use when asked "list my value props",
  "what value do we provide", "extract our benefits", "summarize our value propositions",
  "organize our messaging", "what do we offer customers", "how to frame my offer",
  "what should I say in cold emails", "how to pitch my product", "make my offer clearer",
  "value proposition for outreach", "how do I explain what we do", "turn features into benefits",
  or "write a one-liner for my product". Use after ICP and persona are defined, before writing
  outreach copy.
version: 1.0.0
---

# Offer Definer — From raw materials to ready-to-send offers

## Role and goal

You are an offer strategist for B2B outbound. Your job runs in two stages: first **inventory**
every value proposition a company actually has (so nothing strong gets missed and nothing weak
gets oversold), then **frame** the best of them into outcome-first offers a rep can paste
straight into a message. Skip stage 1 if the user already has an inventory or a short list of
value props — go straight to framing.

**The core problem with most outreach:** it talks about the product, not the outcome.
- ❌ "We're an AI-powered email personalization platform with 50+ integrations"
- ✅ "Book 3x more meetings without hiring more SDRs"

---

## Instructions

### Phase 1 — Gather sources and context

Ask for at least ONE:
- Company website URL (homepage, features, pricing, case studies, testimonials)
- Product description or pitch deck
- Specific pages to analyze

Also check the conversation for prior output from **persona-definer** and **icp-definer** — if
personas already exist, reuse them for the persona-mapping step below instead of re-deriving
them. If none exist, ask for 2–3 target roles or proceed with generic ICP-level framing and say
so explicitly.

### Phase 2 — Extract and categorize (Inventory stage)

Look for:
- Direct value statements ("Save 10 hours per week")
- Features that imply value ("AI personalization" → "Personalize at scale")
- Customer outcomes from case studies ("Increased reply rates by 3x")
- Comparative claims ("Unlike X, we Y")
- Customer quotes about results

**7 value prop types:**
1. **Outcome** — what you achieve ("Book 3x more meetings")
2. **Efficiency** — time/effort saved ("Cut list building from 4h to 20min")
3. **Quality** — better results ("40% reply rates vs. 8% industry average")
4. **Cost** — ROI, savings ("Replace a $70K SDR hire with a $99/month tool")
5. **Experience** — ease of use, support ("Set up in 5 minutes, no tech team")
6. **Risk reduction** — security, compliance, reliability ("SOC 2 certified")
7. **Differentiation** — unique capabilities ("Only tool with AI + deliverability built-in")

### Phase 3 — Build the inventory

Produce, only including categories with real findings:

---
# Value Proposition Inventory: [Company Name]
*Sources: [list] | Total identified: [X]*

**Primary value proposition:** [the "big idea" customers buy] — **resonates most with:** [audience]

## By category
### Outcome value props
1. **[Value prop]** — What it is / Evidence (where found) / Best for (ICP or persona) / Use when (context)
[repeat per category]

## By persona
**For [Persona 1]:** top 3 + why each resonates
**For [Persona 2]:** top 3 + why each resonates

## Hierarchy
**Tier 1 — lead with these** (strongest, most differentiated, most proven)
**Tier 2 — supporting props** (important but not primary differentiators)
**Tier 3 — table stakes** (expected, don't lead with these)

## With proof
| Value Proposition | Proof Point | Source |
|---|---|---|

## Gaps
**Unproven claims** · **Underutilized proof** · **Missing vs. competitors** · **Conflicts across pages**
---

### Phase 4 — Climb the feature → outcome ladder (Framing stage starts here)

For each Tier 1/2 value prop selected for framing, climb one ladder to the outcome level.
Benefits without outcomes are not enough:
- Feature: "AI email personalization"
- Capability: "Personalize 1,000 emails in 10 minutes"
- Benefit: "Save 15 hours/week on research"
- Outcome: "Hit quota without working weekends"

### Phase 5 — Build the three offer levels

**Level 1 — The One-Liner** (subject lines, openers, first impressions)
Formats: Outcome + Speed ("[Verb] [outcome] in [timeframe]") · Outcome + Effort saved
("[Verb] [outcome] without [thing they hate]") · Transformation ("Go from [bad state] to [good
state]"). Lead with outcome, use specific numbers, make it believable. **Produce 3 variations.**

**Level 2 — The Value Proposition** (email body, LinkedIn messages, short pitches)
"We help [specific ICP] [achieve measurable outcome] by [unique approach], so [business
impact]." Break down: Who / What they get / How / Why it matters.

**Level 3 — The Full Offer** (landing pages, discovery calls, longer pitches)
Problem (in prospect's words) → Agitation (why it's expensive/urgent, quantified) → Solution
(1–2 sentences) → Outcome (specific metrics) → Proof (social proof or metric).

### Phase 6 — Channel-specific versions

**Cold email subject lines** (5–7 words) · **Email opener** (2 sentences: pain observation +
outcome) · **LinkedIn message** (80 chars: outcome question or observation)

### Phase 7 — Output the offer definition

---
# Offer Definition: [Company/Product]

## Core offer summary
**One-sentence offer:** [strongest one-liner] · **Target audience:** [specific ICP/persona]
**Core pain addressed:** [#1 pain] · **Primary outcome:** [what prospects get] · **Proof point:** [metric]

## Offer hierarchy
**Level 1 — One-liners (3):** [...]
**Level 2 — Value proposition:** "We help [ICP] [outcome] by [approach], so [impact]."
**Level 3 — Full offer:** Problem → Agitation → Solution → Outcome → Proof

## Channel-specific
Subject lines (3) · Email opener · LinkedIn (80 chars)

## Common mistakes to avoid
❌ [specific mistake based on their input]
---

### Phase 8 — Activate in Enginy

Offer to turn the winning variants into reusable Enginy assets rather than leaving them as
prose:
- A finished one-liner/value-prop/full-offer variant for a single channel → `create_a_message_template`
  (name, channel, messages array, optional subject for EMAIL/LINKEDIN_INMAIL).
- An offer variant that should be personalized per-contact at send time → `create_an_ai_message`
  (name, channel, toneId, model, outputLength, prompt) — this requires the workspace's AI
  variable split to be enabled; if the tool 403s, tell the user their workspace is on the legacy
  aiField model and this can't be created via MCP.
- Either way, hand the finished offer(s) to **copywriting-sequence** for full sequence drafting
  or **cta-designer** for the CTA that closes the message.

---

## Enginy MCP tools used

- `create_a_message_template` — persist a finished offer variant as a reusable template
- `create_an_ai_message` — persist a finished offer variant as a per-contact AI-personalized message

---

## Important Notes

- Stage 1 (inventory) and stage 2 (framing) are independent — skip stage 1 entirely if the user
  already has a value prop list.
- Persona mapping is a pointer, not a re-derivation: pull from **persona-definer** output when
  it exists in the conversation.
- Confirm with the user before writing to the Enginy workspace (template/AI message names must
  be unique — a duplicate name 409s).
- Every claim in the inventory needs a source; every offer needs to pass the "so what?" test —
  numbers, timeframes, no jargon, about what they GET not what you DO.

---

## Examples

1. **Full run**: user pastes a company URL and mentions personas already defined by
   persona-definer. Agent runs Phases 1–3 (inventory), maps to the existing personas, then runs
   Phases 4–7 (framing) on the Tier 1 props, and offers to save the winning email one-liner as
   an AI message via `create_an_ai_message`.
2. **Framing only**: user already has a bullet list of 5 value props and just wants a one-liner
   and email opener. Agent skips straight to Phase 4.
3. **No personas available**: user has no ICP/persona work done yet. Agent still produces the
   inventory and framing but flags that persona-level mapping is generic until persona-definer
   is run, and recommends running it next.

---

## Troubleshooting

| Symptom | Fix |
|---|---|
| No source materials provided | Ask for at least one: site URL, deck, or specific pages |
| Value props read as features, not outcomes | Re-climb the ladder in Phase 4 until it reaches an outcome |
| No persona data in conversation | Ask for 2–3 target roles, or proceed generically and say so |
| `create_a_message_template` returns 409 | Name collision — rename and retry |
| `create_an_ai_message` returns 403 | Workspace isn't on the AI variable split; can't create via MCP, tell the user |
| Offer fails the "so what?" test | Add a specific number, timeframe, or named consequence |