outreach-first-draft-engine— agent skill

When This Skill Applies

• BEFORE writing any prospect-facing outreach copy (emails, LinkedIn, DMs, voice notes, follow-ups)
• BEFORE any copy-checking or quality gate skill runs (this produces the draft; quality gates verify it)
• When an execution plan or pipeline reaches any step requiring written messaging
• When the user asks to draft, write, create, rework, or sharpen any outreach copy
• Load proactively whenever messaging output is expected

Background

This skill exists because of a specific, recurring failure: outreach copy that passes every rule in a quality checklist but doesn't read like a real person with a clear thought wrote it. The failure happens because the drafting process starts with rules and builds sentence-by-sentence against a checklist, producing technically compliant but lifeless copy. The best-performing finalized messages consistently start with one clear idea and flow from that idea as a single thought. The rules come second.

This skill encodes the actual cognitive process for arriving at the right first draft. It is not a set of rules. It is a thinking sequence. Quality gate skills are post-production verification. This skill is the pre-production creative engine.

The Core Problem This Skill Solves

When Claude writes outreach, it defaults to:

1. Loading rules from a quality checklist
2. Constructing each sentence to satisfy a specific rule
3. Assembling sentences into a message
4. Checking the assembled message against the checklist

This produces copy that is mechanically correct but reads like a framework being executed. Each sentence serves a rule instead of serving the message. The result is fragmented, over-optimized, and lifeless.

The correct process is the reverse:

1. Ask one question: what does this prospect need to hear that they don't already know?
2. Write the entire message from that one idea as a flowing thought
3. Read it back as the prospect and ask: would I say yes?
4. THEN check it against the framework and fix anything that's actually broken

This skill enforces that sequence. No exceptions.

The Five-Gate Thinking Sequence

Every piece of outreach copy must pass through these five gates IN ORDER before the user sees it. Do not skip gates. Do not reorder them. Do not start writing until Gate 2 is answered.

Gate 1: The Prospect's Chair

Before writing a single word, answer these questions silently (do not output them to the user):

• Who is this person? What is their actual day-to-day world like?
• What are they worried about RIGHT NOW that keeps them up at night?
• What do they already know about their own situation? (This is what you CANNOT open with.)
• What DON'T they know that would make them stop and think?
• If they got this message while scanning 200 emails on a Tuesday morning, what would make them pause?

The answers to these questions are not the message. They are the lens through which you write the message.

The research inventory rule. Before writing a single word of copy, enumerate every specific, named detail from your research about this person and their situation: numbers, dates, organizational changes, people they interact with, technology names, competitive pressures, deadlines. List them silently. Then write FROM those details, not from the account-level narrative that summarizes them. The account narrative is an abstraction built for internal planning. The specific details are the prospect's lived reality. Every time you reach for a general phrase, stop and check whether a specific detail exists that would make the prospect think "this person is actually paying attention to my world." If the detail exists in your research and you didn't use it, the message is weaker than it should be.

The prospect knows their own world. Never open by narrating their situation back to them. Telling them what they already know is narration. Connecting a dot they haven't connected yet is an insight. The first gets deleted. The second earns the next line.

The prospect knows their own expertise. This is different from knowing their own situation. Situation is "we launched product X last October." Expertise is "I've spent 25 years in this function and I can diagram how this works in my sleep." Never explain a mechanism to someone whose career proves they already understand it. The prospect's career arc (years of experience, previous organizations, functional depth) determines what you can IMPLY versus what you need to SPELL OUT. Gate 1 must produce a clear answer to: "What does this person already understand because of who they are, that I must NOT explain?" Every sentence that explains something the prospect already understands makes the message longer without making it smarter. Replace explanations with the human-level consequence they're watching play out, not the operational abstraction they already own.

Verify attribution before you reference it. If your research surfaced a specific initiative, campaign, program, or public statement, confirm it actually belongs to the prospect's scope of responsibility before weaving it into the message. Referencing someone else's initiative as if it's the prospect's shows sloppy research, not deep research. The specificity that was supposed to build credibility will destroy it.

The prospect's professional background determines how you deliver insight. Not all executives evaluate the same way. A clinician, engineer, or analyst who built their career on evidence wants to see the finding, then decide if the details are worth their time. A relationship-oriented executive can be drawn in by curiosity and pattern teases. Gate 1 must produce a clear answer to: "Does this person evaluate evidence or evaluate relationships?" This determines whether you SHOW the insight in the message body or TEASE it and let curiosity earn the meeting. When in doubt, show. Showing always works. Teasing only works for certain personas.

Classify the forcing function before you write. Every account has a forcing function that creates urgency. But forcing functions split into two categories, and the category determines whether you can lead with it.

PROUD forcing functions are things the prospect would bring up at a conference or reference in a board presentation as progress: a new appointment, market expansion, accreditation, a strategic partnership, a digital transformation initiative. These are safe to lead with.

DEFENSIVE forcing functions are things the prospect would rather not discuss with a stranger: data breaches, lawsuits, settlements, financial losses, investigations, leadership departures under pressure, failed implementations, bad public reviews. Leading with them makes the message read like an attack dressed up as insight. The prospect's instinct is to delete.

The rule: if the forcing function involves organizational failure, embarrassment, legal exposure, or reputational damage, NEVER lead with the event itself. Instead, lead with the INVESTMENT they made in response, which they ARE proud of. Every defensive forcing function has a proud response behind it. A breach leads to a security modernization investment. A settlement leads to a compliance overhaul. Lead with the response, not the wound. The prospect will connect the dots without you pouring salt in it.

The prospect is skeptical by default. They receive hundreds of cold messages. Every one claims to have something valuable. Your message earns attention by demonstrating that you understand their world well enough to say something they haven't already heard from their own team.

Gate 2: The One Idea

Every message has exactly one idea running through it. Not two ideas. Not a setup idea and a payoff idea. One idea, expressed as a flowing thought.

Before writing, state the one idea to yourself in a single sentence. Examples:

• "Commercial teams have already been grinding on the exact numbers the government side is about to get evaluated on, and our CRO has been on that side of it."
• "Where does the front door still reflect yesterday's assumptions, even after the business has moved on?"
• "There's a cutoff point where manual outreach stops driving closure and becomes the expensive option, and finding it makes every downstream decision easier."

If you cannot state the one idea in a single sentence, you don't have a message yet. Stop and think until you do.

Test the idea: Does this idea give the prospect something they don't already have? Does it make them reflect on their own situation differently? If not, find a better idea.

Gate 3: Write It Like You Said It

Write the entire message in one pass as if you're saying it out loud to a peer. Not constructing it. Not assembling it. Saying it.

Rules for this gate:

• The first sentence does exactly one thing. It accomplishes one complete thought. Not two things connected by "and." Not a fact followed by a qualifier. One idea, landed fully, in one sentence. Test: describe what the first sentence does. If you need the word "and" to describe it, the sentence is trying to do too much.
• Write the whole message from the one idea. Every sentence should feel like it flows naturally from the one before it. If you have to force a transition, the sentences don't belong together.
• Do not stop to check rules mid-draft. Write the complete thought first. The rules come at Gate 5.
• Vary sentence length naturally. Some short. Some longer and more layered. The rhythm should feel like speech, not like a checklist being executed.
• Merge consecutive short sentences in the body. When the draft produces two or three short declarative sentences in a row, merge them with commas, "and", "but", or "because" to create the rhythm of someone actually talking. Staccato chains of short sentences read like bullet points disguised as prose. This rule applies to body paragraphs only.
• Use connective tissue. "The reason that matters," "Here's the thing," "That's usually where," "Once that's clear." These are the phrases that make a message feel like one flowing thought instead of five disconnected observations.
• Do not over-explain. If the one idea is clear, the prospect will connect the dots. Trust them. The best copy leaves space for the prospect to fill in their own context.
• Never make the sender the subject of a sentence. The entire message keeps the prospect as the central figure. The most common failure is the "credential line": a sentence like "I work with teams navigating that transition" inserted to establish the sender's authority. This line pattern-matches to every vendor InMail the prospect has ever deleted. If the message needs to establish value, do it through the INSIGHT, not through a credential. Show what you know, not who you are. Prospects respond to what you know.
• Purge rep-language. Certain words reveal you're running a sales motion instead of sharing a genuine observation. Common offenders: "companies like yours," "leaders in your space," "I help organizations," "I've been working with," "teams I partner with." Replace with language a peer would use: "the part I keep thinking about," "a handful of," "the pattern I keep seeing."
• Never describe the solution category. A credential line reveals WHO the sender is. Rep-language reveals the sales MOTION. A product tell reveals WHAT the sender sells. All three break the spell. If a prospect could read a sentence and accurately guess what your company sells, it's a product tell. Rewrite using the human consequence instead of the functional description.
• The CTA must be the natural conclusion of the one idea. The CTA should feel inevitable, not bolted on. If the message is about a structural ceiling, the CTA is about finding whether that ceiling exists in their world. If the message is about two groups working the same problem from different angles, the CTA is about whether a conversation is worth having.

Gate 4: The Prospect Test

Read the entire message back as the prospect. Not sentence by sentence. The whole thing, the way they would read it on their phone between meetings.

Answer these questions honestly:

1. Would I keep reading after the first sentence? If the opener narrates my situation back to me, I'm deleting it. If it tells me something I haven't thought about, I'll read the next line.
2. Do I know why this person is contacting me? Not "they want a meeting." Why ME, why NOW, about WHAT specifically.
3. Do I know what I'd get from a meeting? Not a pitch. Something useful to me whether or not I ever buy anything.
4. Can I say yes without feeling like I'm agreeing to a sales pitch? The CTA should feel like a useful decision for me, not a favor I'm doing for the sender.
5. Does this message have one clear thought, or does it feel like multiple ideas stitched together?
6. Does any sentence make the sender the subject instead of me? That's the moment I pattern-match this to "vendor outreach" and move on.
7. Can I guess what this person sells? If any sentence describes a product category clearly enough that I can identify the company's offering, I know I'm being pitched.
8. Does any sentence explain something I already know because of who I am? If I'm a 25-year operations executive and a sentence walks me through a mechanism I could diagram from memory, the message just talked down to me.
9. Does this message open by reminding me of something painful that happened to my organization? If the first paragraph recounts a breach, a lawsuit, or a financial loss, the message reads like someone building a case against me to sell me something. My guard goes up immediately.

If any answer is no, go back to Gate 3 and rewrite. Do not show the user a draft that fails the Prospect Test.

The batch generation honesty problem. When this gate runs inside a multi-prospect session, there is a documented tendency to apply it with decreasing rigor as volume increases. Contact 1 gets a genuine simulation. By contact 8, the gate becomes a rubber stamp. This is the exact moment the gate matters most. The fix: after ALL copy is drafted, come back and re-run Gate 4 on every piece with fresh eyes.

Gate 5: Rule Check (Light Touch)

NOW, and only now, check the draft against your quality framework. But check it as a READER, not as an auditor. The question is not "does every sentence satisfy a rule?" The question is "does anything in this message break the spell?"

Quick scan for:

• Forbidden words or phrases (leverage, synergy, etc.)
• Em dashes (replace with periods, commas, colons)
• Word count within range for the channel
• CTA ends on a question, not a statement
• Vague metaphors and AI-default phrases (see below)

The AI-ism filter. AI-generated copy has a tell beyond forbidden words: vague metaphorical language that sounds like it means something but actually says nothing. These phrases survive every other gate because they're not technically wrong. They're just empty. Common offenders: "gets heavier," "on the horizon," "looming large," "at the forefront," "navigate the landscape," "in an era of," "the weight of," "carries significant implications," "sends a clear signal," "positions them well," "a critical juncture," "the stakes are high." The test: can you replace the phrase with a specific, concrete statement that says what you actually mean? If yes, always use the specific version. Real people in professional contexts say the actual thing. AI defaults to the metaphor. Every vague metaphor is a missed opportunity to say something concrete.

If the message passes Gates 1-4 and reads like a real person with a clear thought, do NOT rewrite sentences to better satisfy individual rules. The fluidity of the whole message is more important than any single rule. Rules are guardrails, not a construction manual.

Voice Note TL;DL Rule

When drafting a voice note script, also draft the accompanying TL;DL (too long; didn't listen), a 2-3 sentence written summary sent alongside the audio. The TL;DL goes through the same five-gate sequence. It should capture the one idea and the CTA in text form. Sending both gets significantly more replies than voice note alone.

The CTA Discipline

CTAs are the single biggest failure point across outreach copy. Here is what keeps going wrong and how to prevent it.

Failure Pattern 1: The Vague Ask

"Worth a conversation?" About what? Why would I block time for a topic I can't identify?

Fix: The CTA must tell the prospect what the meeting is ABOUT. Not by promising specific deliverables, but by naming the topic or question the conversation would address.

Failure Pattern 2: The Over-Promise

"With the data behind those numbers." Now the sender is committed to showing up with specific data they may not have.

Fix: Frame what the meeting offers in terms of the CONVERSATION, not deliverables.

Failure Pattern 3: The Disconnected CTA

The email body earns the right to ask. Then the CTA asks about something different, or asks so generically that the body's work is wasted.

Fix: The CTA must be the natural conclusion of the one idea from Gate 2.

Failure Pattern 4: The Meeting-As-Favor

"Open to scheduling a brief chat?" The prospect is being asked to do something for the sender. The meeting should feel like something THEY'RE getting.

Fix: Frame the CTA so the prospect is the one who benefits.

Failure Pattern 5: The Credential Bridge

"I work with teams navigating this exact challenge." The message was about THEIR world until this sentence. Now it's about the sender.

Fix: If the CTA needs a bridge, give the insight in the body and let the CTA offer the operational details. Do not tease. Do not withhold. The prospect should get genuine value from reading the message whether or not they reply. The meeting earns its place because they want the specifics.

Default to showing insight, not teasing it. The only time a tease is appropriate is when Gate 1 identifies the prospect as relationship-oriented. Evidence-oriented prospects (engineers, analysts, operations executives) will recognize withholding and resent it.

The Three-Beat Cold Email Close

First cold emails use a specific CTA structure:

1. Proof point sentence: One short sentence about a peer or comparable organization.
2. Value statement: One standalone sentence starting with "Might be helpful/useful/valuable to walk you through..."
3. Time ask (separated by a blank line): "Worth 15 min in the next few weeks?" as its own line.
4. Sign-off (separated by a blank line): First name only.

The three beats create a descending rhythm: context, offer, ask. Merging these into one flowing sentence kills the punch.

Time anchor calibration: "in the next few weeks" for VPs and C-suite. "this week or next" only for directors and mid-level contacts. Tight time anchors for senior executives signal desperation.

What This Skill Does NOT Do

• It does not replace quality gate skills. Quality checkers are still the post-production verification. This skill produces the draft.
• It does not contain formatting rules, word count limits, or channel-specific mechanics. Those belong in your quality gate.
• It does not handle sequencing, scheduling, or touchpoint planning.
• It does not generate cold call scripts or voicemail scripts (it DOES generate voice note TL;DL summaries).
• It does not conduct research or build account briefs. Those are upstream.

This skill does ONE thing: it ensures that the first draft of any written outreach sounds like a real person wrote it in one sitting with one clear idea, and that it would survive being read by a skeptical executive scanning their inbox on a Tuesday morning.

Process for Multi-Message Output

When generating copy for multiple prospects in one session:

1. Run Gates 1-2 for EACH prospect individually before writing anything. Each person gets their own one idea. If two prospects end up with the same one idea, one of them is wrong.
2. Write each message independently through Gates 3-5. Do not batch-write by applying the same template across prospects.
3. After all messages are written, read them back-to-back and verify they sound like different messages written to different people, not the same structure with nouns swapped.

Interaction with Quality Gate Skills

The intended workflow when both this skill and a copy-checking skill are active:

1. First Draft Engine (this skill): Gates 1-4 produce a draft. Gate 5 does a light rule check.
2. Quality Gate Skill: Runs its full checklist and flags anything that needs fixing.
3. Resolution: If the quality gate flags a rule violation but the message reads naturally and passes the Prospect Test, the message wins. If the quality gate flags something that genuinely breaks the spell (forbidden word, wrong sign-off), fix it without disrupting the flow.

The quality gate should NEVER be used as the starting point for writing. It is a verification layer, not a construction tool.

Output

This skill does not produce a separate file. It governs the PROCESS by which outreach copy is drafted. The output is the draft itself, delivered in whatever format the requesting skill or user expects.

When the user asks for copy directly (not as part of a larger plan), deliver:

• The message, ready to send
• One sentence explaining the one idea behind it (so the user can evaluate whether the idea is right before evaluating the words)
• Nothing else unless the user asks for it

About

Built by Dallas Andrews. Part of a 30+ skill AI-powered operating system adopted by five departments without a mandate.