LinkedIn Invites Retargeting for AI Consulting

Retarget linkedin-invites from a UAS and defense-first scoring model to an AI consulting growth model. The skill should prioritize likely buyers, partner-network people, capital sources, AI peers, and warm-path connectors for fast-growth CPG brands, while keeping UAS as a secondary adjacency lane instead of the main lens.

The simplest good version keeps the current browser workflow and approval gates intact, but replaces the scoring rubric with a lane-aware ICP model. That means no new tooling, no new automation risk, and a much better top-of-list ranking for the people Pete actually wants to know now.

1. Problem Statement and Goal

linkedin-invites still behaves like an evolved UAS-era filter. It has some AI automation carve-outs, but the center of gravity is still drones, defense, robotics, and hardware-adjacent technical people.

That creates the wrong sort order. It can overvalue UAS-adjacent engineers or investors while undervaluing:

• heads of People and Culture or operations leaders at 25-150 employee CPG brands who own AI and automation
• AI deployment firms and individuals who could become implementation partners
• investors, operating partners, and portfolio-value teams at VC and PE firms
• connectors who can open doors into the target buyer set
• peers building in the exact AI automation lane Pete wants to be known in

The goal is to make the skill rank invite value based on present business direction, not legacy industry identity.

2. Success Metric

The change is working when all of the following are true:

• in a mixed batch of invitations, buyer, partner, capital, connector, and AI-peer profiles reliably sort above generic UAS-adjacent profiles unless the UAS person is unusually strategic
• invites from fast-growth CPG operators, P&C leaders, IT owners, automation partners, and relevant VC/PE platform people consistently land in ✅ Accept or ⭐ VIP
• generic drone operators, broad agencies, vague AI consultants, and unrelated service founders consistently fall below threshold
• every reviewed invite includes a lane tag so Pete can see *why* the person is valuable, not just the numeric score

3. Current State

Today the workflow in skills/linkedin-invites/SKILL.md:

• gathers pending invites
• researches each profile
• scores each person on Seniority, Company Quality, and Industry Relevance
• applies a note modifier
• classifies as ⭐ VIP, ✅ Accept, or ❌ Decline

The current rubric in skills/linkedin-invites/references/scoring-rubric.md explicitly favors:

• drones, robotics, defense-tech, autonomous systems
• UAS technical leaders and investors
• AI and automation only as a related lane, not the main lane

The workflow itself is fine. The mismatch is almost entirely in the scoring logic and the guidance language around relevance.

4. Platform Capabilities

OpenClaw already supports everything needed for this change with the current setup:

• signed-in browser navigation for LinkedIn review
• read-only profile gathering before approval
• human-in-the-loop approval before any accept, ignore, or message action
• structured plain-text output with ranked recommendations

No new browser capability, no new data source, and no new integration is required. This is an instruction and rubric update inside an existing skill.

5. Community Patterns

The proven pattern for invite triage is not deeper automation. It is tighter ICP scoring plus human review.

For this use case, the useful pattern is:

• separate people by relationship value, not just title prestige
• distinguish buyers from partners, connectors, capital, and peers
• reward direct route-to-market relevance over abstract industry adjacency
• keep final actions gated behind manual approval

That pattern matches Pete's actual use case better than a single "industry relevance" lens.

6. Options

7. Recommendation

Choose Option B.

It preserves the current skill shape, which means less implementation risk, while fixing the real problem: the skill needs to understand *why* someone matters. A single numeric score is not enough when Pete cares about several different kinds of value:

• direct buyer access
• implementation partners
• investor and portfolio access
• connectors and referrers
• AI peers in the target lane
• selective UAS adjacency

Proposed scoring model

Keep the current base score framework:

• Role Fit: 0-4
• Company / Platform Value: 0-3
• Strategic Relevance: 0-3
• Connection Note Bonus: +0 to +2

Add one required lane tag per invite:

• buyer
• partner
• capital
• connector
• ai-peer
• uas-adjacent
• none

Lane definitions

Proposed scoring guidance

Role Fit (0-4)

• 4: founder, CEO, COO, VP, Head, or clear budget / initiative owner in a target lane
• 3: director, senior operator, partnerships leader, portfolio operations lead, solutions lead, or senior technical implementer in a target lane
• 2: strong individual contributor or manager with obvious relevance and likely network value
• 1: generic seller, generic recruiter, vague consultant, or service founder without direct fit
• 0: student, entry-level, unclear profile

Company / Platform Value (0-3)

• 3: fast-growth CPG brand in the target size band, strong AI deployment firm, relevant VC / PE platform, or clearly credible AI product company
• 2: adjacent but promising company, useful mid-market operator, or specialist consultancy with clear implementation depth
• 1: broad service business, vague agency, solo consultant, or tangential company
• 0: unverifiable, spammy, or clearly off-target

Strategic Relevance (0-3)

• 3: direct fit to Pete's current growth thesis, buyer access, partner delivery, or AI automation reputation in the target lane
• 2: useful adjacency, including selected UAS or robotics people, commerce-tech operators, and AI ecosystem people with plausible relationship value
• 1: light relevance but weak route to business value
• 0: unrelated

Classification thresholds

Keep the current thresholds:

• 8-10 = ⭐ VIP
• 4-7 = ✅ Accept
• 0-3 = ❌ Decline

But add one policy change:

• buyer, partner, capital, and strong connector profiles should have an easier path to ⭐ VIP than a generic UAS-adjacent profile

Explicit scoring boosts

The rubric should explicitly boost:

• P&C leaders who clearly own or influence AI automation
• IT leaders at smaller fast-growth brands where IT is folded into ops or people functions
• operations and enablement leaders tasked with process improvement
• AI deployment firms with real implementation depth, not vague "we help companies with AI" positioning
• PE operating partners and portfolio-value teams
• VC partners or principals active in AI, commerce, or operator-heavy investing
• people with credible relationships into CPG brands or commerce operators

Explicit downgrades

The rubric should explicitly downgrade:

• generic agency founders
• vague AI consultants with no delivery proof
• lead-gen shops and no-code spam
• generalist salespeople without domain or route-to-market relevance
• UAS people whose only relevance is legacy familiarity

8. Security Considerations

This change should not loosen any action guardrails.

• no accepting invites without Pete's approval
• no sending messages without Pete's approval
• no CRM enrichment or contact-detail lookup in channel context
• no expansion from profile review into off-platform research unless the person is ⭐ VIP or Pete asks

The main failure mode is ranking drift, not data exposure. If the rubric is too loose, the skill will accept lots of low-value "AI" people. If it stays too UAS-heavy, the business-development value stays capped.

9. Implementation Scope

David should update:

• skills/linkedin-invites/references/scoring-rubric.md
• skills/linkedin-invites/SKILL.md

Likely changes:

• rewrite the "Wants to connect with" and "Not interested in" sections
• add the lane-tag requirement
• rewrite score examples around CPG buyers, partner-network people, capital, connectors, and AI peers
• move UAS from a core scoring lane to an adjacency lane
• add explicit examples for P&C, IT-owner, operations, AI deployment partner, VC, and PE roles
• update result presentation so each invite shows the lane tag in the ranked output

No browser or logging changes are required unless we also want lane tags written into linkedin_log.ndjson.

10. Validation Criteria

Pete should be able to test the updated skill against example profiles like these:

1. Head of People at a 90-person CPG brand who owns AI tooling evaluation

Expected: buyer, ✅ Accept or ⭐ VIP

1. Founder of a specialist AI automation deployment firm with real case evidence

Expected: partner, ✅ Accept or ⭐ VIP

1. PE operating partner focused on portfolio efficiency and AI adoption

Expected: capital, ✅ Accept or ⭐ VIP

1. Commerce-tech connector with strong CPG network and clear intros value

Expected: connector, ✅ Accept

1. AI solutions leader at a credible automation platform selling into operating teams

Expected: ai-peer or partner, ✅ Accept or ⭐ VIP

1. UAS engineer with strong product depth but no direct route to current business goals

Expected: uas-adjacent, usually ✅ Accept, not automatic ⭐ VIP

1. Drone services founder with vague AI claims

Expected: none or uas-adjacent, ❌ Decline

1. Generic "AI consultant" with buzzwords and no implementation proof

Expected: none, ❌ Decline

11. Category

Skill

This is a behavior change inside an existing multi-step workflow skill.

12. Context Loading

On every linkedin-invites run, load:

• skills/linkedin-invites/SKILL.md
• skills/linkedin-invites/references/scoring-rubric.md

Load only when drafting or sending a message:

• skills/linkedin-invites/references/writing-style.md

Do not load personal memory files for this workflow in group-chat contexts.

13. Guardrails

The updated skill must not:

• auto-privilege UAS or defense profiles just because they match Pete's history
• auto-score founders highly when the company is actually a generic service shop
• treat vague "AI advisor" language as evidence of partner quality
• accept company page follow requests
• send custom messages to non-VIP accepts
• use channel context to surface personal contact data

The updated skill should prefer present business direction over nostalgia.

14. Handoff

Before implementation, Vinny should hand Pete:

• this spec artifact for review
• a short summary of the proposed lane model
• the key policy shift: UAS remains in play, but as a secondary lane

After Pete approves, David should update the skill files. Vinny should then run a fresh li invites scan and verify that:

• lane tags appear in output
• top-ranked invites look commercially right
• the accept / VIP / decline mix feels aligned with current business goals