Holistics AI Skills

Coming Soon

AI Skills is coming soon. Reach out to us via [email protected] if you're interested in early access.

Overview

If you've used custom skills in AI assistants (like Claude, ChatGPT, or Copilot), you'll get the idea. AI Skills brings the same concept to Holistics AI: instead of cramming everything into a single context.aml file, Analysts can define multiple skill definitions right in AML, each scoped to specific teams with its own instructions, data references, and access controls.

Key capabilities

• Domain-specific skills: create a finance_skill and an ecommerce_skill with different instructions, terminology, and dataset references. Each team gets AI responses tailored to their work.
• Per-user access control: use user attribute expressions to restrict skills by department, role, or any attribute (e.g., disabled: user.department != 'Finance').
• Flexible activation modes: skills can auto-activate when relevant, require manual invocation via slash commands, or stay hidden as background context. Six modes are available: auto, manual, always-on, hidden, disabled, and switchable variants.
• Dynamic context references: inject datasets, contexts, or files with preloaded (${}) or lazy (@Context:, @Dataset:, @File:) references. Lazy references keep the AI's context window lean until content is actually needed.
• Import from existing docs: upload Claude or markdown skill files and convert them into AML skill definitions.

How it works

Skills are defined as an AML type in the data model:

Skill finance_analysis {
  label: "Finance Analysis"
  disabled: H.current_user.team != "Finance"
  description: @md Specialized AI skill for finance team queries;;
  content: @md Use ${finance_metrics} for all queries;;
}

End-users interact with skills through slash commands (/finance_analysis), UI toggles, or simply by chatting. The AI auto-selects relevant skills by default. Analysts control who sees which skills and how they are activated.

Use cases

• Domain specialization: finance, ecommerce, and support teams each get skills with domain-specific instructions and data references, instead of sharing one generic context.
• Standardized reporting: a client_performance_report skill defines exactly which KPIs to pull, how to group them, and what benchmarks to compare against, so every team member gets consistent output.
• Multi-step analysis: a quarterly_business_review skill guides the AI through a sequence: pull revenue by region, compare to targets, flag underperformers, and cross-reference with marketing spend.
• Data quality checks: a data_quality_audit skill runs a series of checks (missing IDs, duplicates, threshold breaches) and produces a structured summary with a health score.