Skip to main content

Skills

Skills are specialized knowledge packages that give agents domain expertise. Think of the difference between tools and skills this way: tools let an agent do things (call an API, send an email), while skills teach an agent how to think about a domain (sales methodology, incident triage, cloud architecture).

How Skills Work

Skills follow the Anthropic Agent Skills specification. They are markdown-based knowledge bundles that agents load on-demand when they encounter a relevant situation.

For example, when a sales agent encounters a pricing objection, it loads the sales-objection-handling skill and gets the LAER framework, response scripts, and competitor battlecards -- all without that knowledge taking up space in every single conversation.

The Three-Level System (Progressive Disclosure)

Skills use progressive disclosure to minimize context usage:

LevelWhat LoadsWhenToken Cost
Level 1Name and short description onlyAlways (in system prompt)~100 tokens
Level 2Full instructions, frameworks, templatesWhen the agent activates the skillUp to ~5,000 tokens
Level 3Detailed reference files (guides, battlecards, examples)When the agent needs specific depthOn demand

This means agents carry awareness of all their skills at minimal cost, and only load the heavy content when they actually need it.

Skills vs Tools

AspectToolsSkills
PurposeDo things (API calls, calculations)Know things (expertise, frameworks)
OutputAction results (data, side effects)Guidance (instructions, templates)
LoadingAlways availableLoaded on-demand (progressive disclosure)
FormatAPI integrationsMarkdown + reference files

Available Skills

MeetLoyd ships with 110+ first-party skills across these categories:

Engineering and Cloud

SkillDescription
coding-agentFull-stack development with sandboxed execution
aws-architectAWS Solutions Architect -- Well-Architected Framework, IaC, cost optimization
azure-architectAzure Solutions Architect -- Bicep, Entra ID, AKS
gcp-architectGCP Solutions Architect -- Cloud Run, GKE, BigQuery, Terraform
multi-cloud-architectMulti-cloud and hybrid -- Kubernetes, Terraform, vendor lock-in mitigation
terraformTerraform/OpenTofu IaC, modules, state management, CI/CD
kubernetesDeployments, Helm charts, RBAC, networking, troubleshooting
postgresqlQuery optimization, indexing, replication, PL/pgSQL
kafkaEvent streaming, Kafka Streams, Connect, Schema Registry

Enterprise IT

SkillDescription
servicenowITSM/ITOM, AI Agents, Now Assist, CMDB
oktaIdentity, SSO/SAML, OAuth 2.0, MFA, Zero Trust
datadogObservability, APM, logs, metrics, Bits AI
pagerdutyIncident management, SRE Agent, on-call, escalations
splunkLog management, SPL queries, SIEM, threat detection
grafanaDashboards, PromQL, Loki, Tempo, alerting
vaultHashiCorp Vault secrets management, dynamic credentials, PKI

Sales and Marketing

SkillDescription
sales-objection-handlingB2B SaaS objections with LAER framework, 25+ patterns
strategic-narrativeAndy Raskin framework, positioning, category creation
growth-hackingViral loops, experimentation, PLG, AARRR metrics
inbound-marketingContent strategy, SEO, lead nurturing, automation
google-adsSearch, Display, Performance Max, Smart Bidding
linkedin-adsABM campaigns, Lead Gen Forms, B2B targeting
content-managementLinkedIn, Twitter/X, blog, content calendar
pr-communicationsPress releases, media relations, crisis comms

CRM and Billing

SkillDescription
salesforceApex, LWC, SOQL, Einstein AI, Flow
hubspotCRM/Marketing APIs, workflows, Breeze AI
stripePayments, subscriptions, Connect, Checkout
zuoraSubscriptions, revenue recognition (ASC 606), quote-to-cash

SaaS Business

SkillDescription
saas-metricsLTV/CAC, NRR, Rule of 40, cohort analysis
saas-pricingValue-based pricing, freemium, packaging
saas-valuationARR multiples, DCF, comps, fundraising benchmarks

Web Development

SkillDescription
laravelLaravel PHP (11/12), Eloquent, Blade, queues, FrankenPHP
drupalCMS development, modules, themes, Twig, Entity API
dsfrFrench Government Design System (RGAA compliance)

AI Development

SkillDescription
openai-agentsOpenAI Agents SDK, multi-agent systems, handoffs
crewaiCrewAI orchestration, agent crews, task delegation
anthropic-mcpMCP server development, tools, resources
ml-trainingPyTorch, fine-tuning, hyperparameter optimization
llm-finetuningLoRA, QLoRA, DPO, ORPO, evaluation strategies

Productivity

SkillDescription
spreadsheetsExcel and Sheets -- XLOOKUP, QUERY, pivot tables, Apps Script
presentationsSlides and PowerPoint -- deck design, data viz, narrative
jiraJQL queries, Atlassian Intelligence, Rovo AI, automation
indesignAdobe InDesign -- enterprise PDFs, whitepapers, reports

Skill Discovery

Agents can browse the skill catalog and request skills they need. Skills are also suggested automatically when integrations are connected or during team onboarding. See Skill Discovery for details.

Assigning Skills

Via the Dashboard

The easiest way to manage agent skills:

  1. Open Agent Settings (edit an agent)
  2. Scroll to the Skills section
  3. Click Assign Skill to browse available skills
  4. Click a skill to assign it
  5. Use the toggle to enable/disable a skill without removing it
  6. Click the delete icon to unassign a skill completely

The skills section shows each assigned skill with its category, description, and enabled state.

Via Loyd

Ask Loyd:

"Give the sales agent the objection-handling skill"

From the Store

  1. Go to Store > Skills
  2. Browse or search for skills
  3. Click Subscribe (skills are subscription-based)
  4. Assign to agents from the dashboard

Creating Custom Skills

Skill Structure

A custom skill follows this directory layout:

my-skill/
├── SKILL.md # Required: frontmatter + instructions
├── references/ # Optional: detailed reference docs
│ ├── GUIDE.md
│ └── EXAMPLES.md
└── assets/ # Optional: images, diagrams

SKILL.md Format

The SKILL.md file has YAML frontmatter followed by the skill content:

Frontmatter FieldDescription
nameUnique skill identifier
descriptionWhat the skill does (max 1024 chars) -- this is the Level 1 text
licenseLicense type (e.g., MIT)
metadata.authorAuthor name
metadata.versionSemantic version
metadata.categoryCategory for browsing
metadata.tagsKeywords for search

The body of the file contains:

  • When to Use section -- tells the agent when to activate the skill (and when NOT to)
  • Main Content -- instructions, frameworks, templates, decision trees
  • Reference Files -- links to detailed reference documents in the references/ directory

Best Practices for Custom Skills

PracticeWhy
Clear triggersTell the agent exactly when to activate and when NOT to ("Activate when customer mentions pricing. Do NOT activate for general product questions.")
Structured contentUse tables, decision trees, and templates -- not walls of text
Actionable templatesProvide copy-paste response templates with placeholders
Reference files for depthKeep main instructions concise, put detailed content in references

Skill Discovery
Auto-discovery and connector-based suggestions
Coding Agent
Give agents development capabilities
Creating Skills
Build and publish your own skills