10 Real AI Skills from Production Deployments

Theory is useful. Examples are actionable.

These are real Skills-as-Documents artifacts drawn from NimbleBrain production deployments and our own internal operations. Company names and proprietary details are anonymized, but the structure, logic, and judgment patterns are authentic. Each example shows the complete skill markdown and explains how an AI agent uses it in practice.

Copy any of these. Modify the criteria for your business. Deploy today.

1. Lead Qualification

The skill most organizations build first. Replaces the mental model a senior sales rep applies to every inbound lead.

# Lead Qualification Skill

## Purpose
Score inbound leads (1-100) and assign routing tier.

## Inputs
- Company name, industry, employee count
- Lead source (inbound, referral, outbound, event)
- Stated pain point or intended use case
- Current tech stack (if known)

## Criteria
- Employee count 50-500: +20 points
- Industry in [SaaS, fintech, healthcare, manufacturing]: +15
- Inbound or referral source: +25 / Event: +15 / Outbound: +5
- Pain point references AI, automation, or operational efficiency: +20
- Has existing AI or automation investment: +10
- Previously engaged (demo, webinar, content download): +10

## Decision Rules
- Score >= 70: HOT, route to founder for same-day outreach
- Score 40-69: WARM, enroll in nurture sequence, review weekly
- Score < 40: COLD, add to newsletter, no active outreach

## Exceptions
- Any company > 500 employees: always HOT
- Referral from existing client: always HOT regardless of score
- Government or defense sector: route to founder regardless of score

How the agent uses it: The agent reads each inbound lead’s submission, gathers the input data, applies the scoring criteria, and assigns a tier. The exceptions section prevents misroutes: an enterprise referral never lands in the newsletter. One client reduced lead response time from 6 hours to 12 minutes by deploying this skill on their inbound pipeline.

2. Content Review

Encodes editorial judgment so content quality stays consistent regardless of who (or what) writes the first draft.

# Content Review Skill

## Purpose
Evaluate draft content for publication readiness. Return a score
and specific feedback.

## Inputs
- Draft content (full text)
- Target audience (practitioners, executives, technical)
- Publication channel (blog, email newsletter, social, docs)

## Criteria
- Every claim backed by a specific number, timeline, or outcome: +25
- Active voice throughout, no passive constructions: +15
- No banned phrases (leverage, synergy, journey, cutting-edge,
  best-in-class, stakeholder alignment): +15
- Length appropriate for channel (blog: 800-1500 words,
  email: 200-400, social: under 280 characters): +10
- Clear, specific call to action: +10
- Headline uses contrast or specificity, not generic benefit: +15
- Paragraphs are 2-4 sentences max: +10

## Decision Rules
- Score >= 80: PUBLISH, ready for final proofread
- Score 50-79: REVISE, return with specific line-level feedback
- Score < 50: REWRITE, structural issues, needs new approach

## Exceptions
- CEO or founder-authored content: flag concerns in comments,
  do not block publication
- Time-sensitive announcements (product launch, incident response):
  lower PUBLISH threshold to 60

How the agent uses it: The agent reads the draft, evaluates each criterion, and returns a structured assessment with the overall score, per-criterion breakdown, and line-specific feedback for anything scoring below threshold. This is a real skill from NimbleBrain’s own operations: every blog post and email draft runs through it before a human sees it.

3. Customer Escalation Routing

Determines when a support interaction needs to leave the normal queue and who should handle it.

# Escalation Routing Skill

## Purpose
Determine if a customer interaction requires escalation and
route to the appropriate handler.

## Inputs
- Customer tier (enterprise, growth, starter)
- Issue category (technical, billing, account, feature request)
- Customer sentiment (positive, neutral, frustrated, angry)
- Days since last interaction on same issue
- Whether the customer has mentioned cancellation or alternatives

## Criteria
- Sentiment is angry or frustrated: +30 escalation points
- Customer mentioned cancellation, "looking at alternatives,"
  or "considering switching": +40 escalation points
- Enterprise tier: +20 escalation points
- Same issue reopened within 7 days: +25 escalation points
- Issue unresolved for more than 3 interactions: +20 escalation points

## Decision Rules
- Escalation score >= 60: ESCALATE, route to customer success
  manager with full context summary
- Escalation score 30-59: FLAG, add internal note, keep in
  standard queue but notify team lead
- Escalation score < 30: STANDARD, handle normally

## Exceptions
- Any mention of legal action or regulatory complaint: immediate
  escalate to VP of customer success, regardless of score
- Customer within 30 days of renewal: add +15 escalation points
  and CC account executive on any interaction
- VIP list customers (defined in context): always escalate

How the agent uses it: The agent evaluates every customer interaction in real time, scoring for escalation signals. The exceptions section is where this skill earns its value: catching the renewal-window churn risk or the legal mention that a junior support rep might miss. One deployment caught 23 at-risk renewals in the first month that would have gone unnoticed in the standard queue.

4. Pricing Decision

Encodes the judgment a sales leader applies when deciding whether to offer a discount and how much.

# Pricing Decision Skill

## Purpose
Evaluate discount requests and determine appropriate pricing
adjustments.

## Inputs
- Standard price for requested package
- Customer's stated budget or price concern
- Deal size (annual contract value)
- Customer tier and history
- Competitive situation (is customer evaluating alternatives?)

## Criteria
- Annual contract value > $100K: eligible for volume discount (up to 15%)
- Multi-year commitment (2+ years): eligible for term discount (up to 10%)
- Competitive displacement (switching from named competitor): eligible
  for migration discount (up to 20%)
- Strategic account (reference-able brand, case study potential):
  eligible for partnership discount (up to 10%)
- Discounts are additive but capped at 30% total

## Decision Rules
- Total eligible discount <= 15%: sales rep can approve
- Total eligible discount 16-25%: requires sales director approval
- Total eligible discount 26-30%: requires VP approval
- Any request exceeding 30%: decline with counter-offer at 30%

## Exceptions
- First 3 customers in a new vertical: founder can approve up
  to 40% as market entry pricing
- Renewals with expansion: no discount on expansion ARR, existing
  discount grandfathered
- Non-profit or education: flat 25% discount, no stacking

How the agent uses it: The agent evaluates a discount request, calculates eligible adjustments based on the criteria, determines the total, and routes to the appropriate approver. The skill prevents unauthorized discounting while preserving flexibility for strategic situations. The exceptions encode institutional knowledge (like the market entry pricing authority) that would otherwise live in the founder’s head.

5. New Customer Onboarding Checklist

Transforms a checklist that lives in a project management tool into an executable skill that an agent can drive.

# Customer Onboarding Skill

## Purpose
Execute the onboarding sequence for new customers and track
completion.

## Inputs
- Customer name and primary contact
- Purchased package and features
- Technical environment (cloud provider, key integrations)
- Stated goals for first 90 days

## Sequence
1. Send welcome email with onboarding timeline (Day 0)
2. Schedule kickoff call within 3 business days (Day 0-1)
3. Provision account and configure purchased features (Day 1-2)
4. Send pre-kickoff questionnaire (Day 1)
5. Conduct kickoff call: align on goals, introduce team (Day 3-5)
6. Complete initial configuration based on questionnaire (Day 5-7)
7. Schedule training sessions (Day 7-10)
8. Conduct training (Day 10-15)
9. First health check: validate adoption metrics (Day 30)
10. Second health check, review against stated goals (Day 60)

## Escalation Triggers
- Any step overdue by more than 2 business days: notify CS manager
- Customer unresponsive for more than 5 business days: escalate to
  account executive
- Kickoff call not scheduled within 5 business days: escalate to
  VP of customer success

How the agent uses it: The agent treats the onboarding skill as a stateful workflow. It tracks which steps are complete, triggers the next action when conditions are met, and escalates when timelines slip. The skill replaced a Notion checklist that was manually tracked. Now the agent drives the process and alerts humans only when intervention is needed.

6. Vendor Evaluation

Standardizes how the team evaluates vendor proposals so decisions are consistent and auditable.

# Vendor Evaluation Skill

## Purpose
Score vendor proposals against standard criteria for
procurement decisions.

## Inputs
- Vendor name and proposal document
- Product/service category
- Annual cost
- Contract terms (length, cancellation, SLA)

## Criteria
- SOC 2 Type II certification: required (disqualify if absent)
- Uptime SLA >= 99.9%: +20 points
- Data residency options (US, EU): +15 points
- API-first architecture with documented endpoints: +20 points
- Annual cost within budget allocation: +15 points
- References from similar-sized companies: +15 points
- Contract allows termination with <= 90 days notice: +15 points

## Decision Rules
- Score >= 80 and no disqualifiers: APPROVE, proceed to contract
- Score 60-79: CONDITIONAL, request remediation on gaps
- Score < 60 or any disqualifier: REJECT, document reasoning

## Exceptions
- Sole-source vendors (no competitive alternative): skip scoring,
  require VP approval with documented justification
- Open-source alternatives available: must evaluate OSS option
  before proceeding with paid vendor

How the agent uses it: The agent reads vendor proposals, extracts the relevant data points, scores against the criteria, and produces a structured evaluation. The SOC 2 requirement as a hard disqualifier prevents wasted time evaluating vendors that can’t meet baseline security standards. This skill made vendor evaluation 3x faster by eliminating the back-and-forth of subjective assessments.

7. Meeting Notes Triage

Determines what happens with meeting notes: which items become tasks, which are informational, which need follow-up.

# Meeting Notes Triage Skill

## Purpose
Process meeting notes and extract actionable items with
owners and deadlines.

## Inputs
- Raw meeting notes or transcript
- Meeting type (internal, client, sales, planning)
- Attendees and their roles

## Extraction Rules
- Any sentence containing "will," "should," "needs to," or
  "action item": extract as a task
- Any sentence with a person's name + a verb: candidate for
  task assignment
- Any question left unanswered: extract as follow-up item
- Any decision stated ("we decided," "agreed to," "going with"):
  extract as decision record

## Output Format
- Tasks: [owner] ([task description]) [deadline if stated,
  else "TBD"]
- Decisions: [decision] ([date]) [participants who agreed]
- Follow-ups: [question/topic]: [suggested owner]
- Key takeaways: 3-5 bullet summary

## Exceptions
- Client meetings: always generate a follow-up email draft
  summarizing decisions and next steps
- Planning meetings: tasks default to sprint backlog format
  with story point estimates

How the agent uses it: After every meeting, the agent processes the transcript, extracts structured items, and routes them to the appropriate systems (tasks to the project tracker, decisions to the decision log, follow-up emails to drafts. The extraction rules aren’t exhaustive) they give the agent patterns to look for while leaving room for judgment on ambiguous items.

Building Your Skills Library

These seven examples cover the most common starting points: sales, content, support, pricing, onboarding, procurement, and operations. But every organization has decisions specific to their domain.

The pattern is universal. Pick a decision. Interview the expert who makes it best. Write down the criteria, rules, and exceptions. Test against real scenarios. Commit to git.

NimbleBrain’s own skills library follows this exact pattern. We use structured markdown skills for proposal generation, code review standards, client communication guidelines, and deployment checklists. Every skill is a document in a repository. Every iteration is a commit. Every improvement compounds across every agent that references it.

That’s Business-as-Code at scale. A library of documents (skip the platform, skip the model) that make your entire AI operation smarter, one skill at a time.

Start with the decision your team makes most often. Write the skill. Test it. Ship it. Then write the next one. By skill number five, you’ll wonder how you ever operated without them.