Six Claude Code Skills for Signal-Based B2B Outbound

Source: github.com/julienamorgan/signal-prospecting-kit

License: MIT

Six Claude Code skills for signal-based B2B outbound. Each skill writes context the next one reads. The system gets sharper with every batch.

The kit does three things:

1. Surfaces buying signals in your market
2. Captures your ICP and brand voice
3. Finds prospects showing those signals and drafts personalised outreach for each

Why signal-based outbound

Outbound works on two axes: fit and timing.

Fit is your ICP. Timing is what is happening right now that makes a prospect more likely to act.

Most teams score on fit alone. The ones that add timing get to prospects before competitors know there is an opportunity.

This kit detects both. It identifies the buying signals that matter for your specific market, finds the companies showing those signals right now, and writes outreach that references something real from each signal.

The six commands

/start          Profile your company, ICP, and won deals
/profile        Research competitors and find white space
/signal-scout   Identify the buying signals that matter for your market
/prospect       Find companies showing those signals right now
/outreach       Draft personalised cold emails and LinkedIn messages
/push           Export to Instantly, Apollo, Outreach, HubSpot, or CSV

You approve at every gate. Nothing moves forward without you. After each approval, the next step starts automatically. No commands to remember.

Install: Claude Code (CLI)

git clone https://github.com/julienamorgan/signal-prospecting-kit.git
cd signal-prospecting-kit
claude

The skills live in .claude/skills/ so Claude Code picks them up automatically. No install script. No restart.

Install: Claude Co-work (Desktop or Web)

1. Download the ZIP: github.com/julienamorgan/signal-prospecting-kit/archive/refs/heads/main.zip
2. Unzip it and open the skills/ folder (all six skills are inside)
3. In Claude: Customize > Skills > + > drag the skill folders in
4. Toggle them on, start a new chat, type /start

The skills/ folder and .claude/skills/ folder contain the same skills. skills/ is the visible one for easy access.

Tools: optional

The kit works with zero tools connected. Web research handles everything. It pattern-guesses emails from public sources and falls back to LinkedIn connection requests when no email is found.

If you have existing sales tools, the kit detects and uses them automatically. Connect via MCP server or API key. Tell the kit what you have during /start and it walks you through connecting it.

The learning loop

After each batch, tell the system who replied. It tracks which signals, openers, and CTAs performed. By batch four, the patterns are clear and the system applies them automatically.

Every other skill kit starts from zero every time. This one does not.

Skill 1 of 6. /start

Run this once when you set up the kit. Profile your company, ICP, and won deals. Every other skill reads what /start writes.

What it does

/start captures the context every other skill needs:

• Your company: what you sell, how you position it, what you do better than alternatives
• Your ICP: titles, company types, sizes, geographies, qualification criteria
• Your won deals: what they had in common, what triggered the buying decision, what the champion said
• Your brand voice: how you write, what sounds like you, what to avoid

It writes this to a context file that every subsequent skill reads. You do not re-explain your ICP to /signal-scout or /outreach. /start handles it once.

Copy-paste prompt

/start

Company: [COMPANY NAME]
What we sell: [one sentence]
Who we sell to: [ICP titles, company types, sizes]
What we do better than alternatives: [one or two specific things]

Won deal patterns:
- Common title: [most common buyer title in won deals]
- Common trigger: [what usually prompted them to look for a solution]
- Common objection we beat: [the one we win on most often]

Brand voice:
- Tone: [direct / consultative / technical / casual]
- What sounds like us: [specific phrases or style notes]
- What to never write: [corporate language, banned words, em dashes]

If anything is unclear, ask me one question at a time before proceeding.

Skill 2 of 6. /profile

Research competitors and find white space. Tells you what your competitors are saying, where they are weak, and which buyer segments nobody is serving well.

What it does

/profile runs competitive research using web search and any tools you have connected.

It looks at:

• Competitor positioning: how they describe themselves, what they lead with, what they avoid
• Customer complaints: G2 and Capterra reviews for the weaknesses customers name in their own words
• White space: segments, use cases, or buyer types nobody in your market is addressing clearly

The output feeds into /signal-scout. The signals worth tracking are often the ones that map to your white space, not the obvious ones your competitors are already targeting.

Copy-paste prompt

/profile

Main competitors to research: [list 2 to 4 competitors]
What I want to know: [positioning gaps / customer complaints / underserved segments / all three]

For each competitor return:
1. How they position themselves in their own words (homepage and pricing page)
2. The 3 most common customer complaints in exact review language from G2 or Capterra
3. The buyer type or use case they are clearly not serving

After the competitive research:
- The one positioning claim I can make that none of them can credibly make back
- The segment most worth going after based on the gaps you found
- Which buying signals would indicate a company in that segment is actively looking

Skill 3 of 6. /signal-scout

Identify the buying signals that matter for your specific market. Not generic intent data. Signals specific to what triggers your buyers.

What it does

/signal-scout takes the ICP and competitive research from /start and /profile and identifies the specific signals that reliably precede a buying decision for your offer.

Signals are not just funding rounds and job postings. The most reliable signals are specific to your market:

• A company hiring a specific title that always precedes budget opening
• A regulatory change that creates urgency in a specific sector
• A product launch that signals a company is moving into a new market
• A leadership change that resets priorities and creates a buying window
• A competitor customer complaint that creates a switching moment

/signal-scout identifies which of these are most predictive for your ICP and writes them as scoreable signal definitions for /prospect to use.

Copy-paste prompt

/signal-scout

Using the ICP and competitive research from /start and /profile:

1. Identify the 5 to 8 signals most predictive of a buying decision for my offer
   For each signal:
   - What the signal is in plain terms
   - Where to detect it (LinkedIn, G2, job boards, news, SEC filings, product pages)
   - How reliable it is as a predictor (high / medium / low) and why
   - How long the buying window stays open after the signal appears
   - The exact opening angle to use when the signal is the reason for reaching out

2. Rank the signals by: reliability x window length x ease of detection

3. For the top 3 signals: write the exact search query or detection method
   to find companies showing that signal right now

4. Write the signal definitions as scoreable criteria for /prospect to use

Skill 4 of 6. /prospect

Find companies showing your signals right now. Not a static list. Live detection against the signal definitions from /signal-scout.

What it does

/prospect uses the signal definitions from /signal-scout to find companies currently showing those signals.

For each company found it returns:

• Company name, size, and relevant details
• Which signal triggered the match and the specific evidence
• The signal recency (how recent is the trigger)
• The recommended contact: title, why this person, and how to find them
• An ICP score: does this company fit on firmographic criteria in addition to the signal

If you have tools connected (Apollo, ZoomInfo, Clay, LinkedIn Sales Navigator), /prospect uses them. If not, it uses web research and pattern-guesses contact details from public sources.

Copy-paste prompt

/prospect

Using the signal definitions from /signal-scout:

Find [N] companies currently showing at least one of the top 3 signals.

For each company return:
1. Company name, industry, size, and HQ location
2. The signal that triggered the match
3. The specific evidence of the signal with source and date
4. How long ago the signal appeared
5. Recommended contact: title, why this person for this signal
6. ICP fit score (1 to 10) against the criteria from /start
7. Anything that would disqualify this company immediately

Rank the list by: signal recency x ICP score

Flag any company where the signal evidence is weak or the ICP fit is below 6.
I approve before anything moves to /outreach.

Skill 5 of 6. /outreach

Draft personalised cold emails and LinkedIn messages for each approved prospect. Every message references the specific signal that triggered the match. Nothing generic.

What it does

/outreach takes the approved prospect list from /prospect and writes a personalised cold email and LinkedIn message for each one.

Every message:

• References the specific signal that triggered the match
• Uses your brand voice from /start
• Matches the tone to the persona (CEO vs VP vs manager)
• Has a specific, low-friction CTA
• Passes a self-check: could this have been sent to anyone? If yes, rewrite.

The learning loop: after each batch, tell /outreach who replied and what they said. It updates the patterns it applies to future batches. By batch four, the system knows which signal-opener combinations work for your ICP and applies them automatically.

Copy-paste prompt

/outreach

Using the approved prospect list from /prospect and the brand voice from /start:

For each prospect write:

1. Cold email
   - Subject line: under 7 words, specific, no corporate language
   - Opening line: references the specific signal, not a generic opener
   - Body: under 75 words, one pain point, one outcome, no feature list
   - CTA: specific and low friction. Not "book a call". Something they can say yes to in one sentence.
   - Self-check: could this email have been sent to any company? If yes, rewrite until it could not.

2. LinkedIn message
   - Under 150 characters
   - References the signal or something specific about them
   - No pitch
   - Opens a conversation, does not close one

Rules:
- No em dashes anywhere
- No corporate language: leverage, synergy, best-in-class
- Sound like the brand voice in /start
- One CTA only. Never two.

The learning loop prompt

Run this after each batch to feed reply data back into the system:

/outreach update

Batch results:
- Sent: [N emails]
- Replied: [N] ([list which signals triggered the replies])
- Positive replies: [N] ([which openers, which signals])
- No reply: [N] ([which signals, which openers])

Update the signal weights, opener patterns, and CTA performance based on this data.
Apply the updated patterns to the next batch automatically.
Tell me what changed and why.

Skill 6 of 6. /push

Export approved outreach to Instantly, Apollo, Outreach, HubSpot, or CSV. One command. You approve the list, /push handles the rest.

What it does

/push takes the approved outreach from /outreach and exports it to your sequencing or CRM tool.

Supported exports:

• Instantly: creates campaign with contacts, sequences, and personalised first lines
• Apollo: creates sequence, adds contacts, loads personalised openers
• Outreach: creates sequence, adds prospects, maps fields
• HubSpot: creates contacts, adds to sequence, sets signal as a note
• CSV: clean export for any other tool

If no tool is connected, /push outputs a clean CSV you can import anywhere.

Every export includes the signal context as a note or tag so the rep who picks up a reply knows exactly what triggered the outreach.

Copy-paste prompt

/push

Export the approved outreach from /outreach.

Destination: [Instantly / Apollo / Outreach / HubSpot / CSV]

For each contact include:
- First name, last name, company, title, email (if found), LinkedIn URL
- The personalised email subject line and body
- The personalised LinkedIn message
- The signal that triggered this contact as a note or tag
- Signal recency as a date field

If exporting to a sequencing tool:
- Set the personalised first line as the opener of touch 1
- Use the generic follow-up template for touches 2 and 3
- Flag any contact where no verified email was found so I can review before sending

If no tool is connected: output a clean CSV with all fields clearly labelled.

Running the full sequence

The complete sequence from setup to sent:

/start     [once on setup, then update when ICP or positioning changes]
/profile   [once on setup, then re-run when a new competitor enters the market]
/signal-scout   [once on setup, then re-run quarterly]
/prospect  [weekly or per batch]
/outreach  [per batch, after approving /prospect list]
/push      [per batch, after approving /outreach output]

After each /push, run the /outreach update with reply data. By batch four the system is applying your reply patterns automatically.

AI Driven, Personalized Outreach on LinkedIn: Prosp 💜

↳ Start a 14 Day Free Trial 💜 ↳ Watch Full Tutorial ▶️