Skills & Pipelines

Skills are procedural instruction files that teach agents how to perform specific tasks. Pipelines chain multiple skills into declarative YAML workflows for complex multi-step operations.

Skills Overview

AI OS includes 19 built-in skill files stored in .claude/skills/. Each skill is a structured Markdown document that an agent reads and follows step by step.

Skill Categories

• Research — Deep research, competitive analysis, literature review
• Content — Blog writing, social posts, email sequences, report generation
• Technical — Code generation, debugging, API integration, database design
• Design — Brand cloning, design system creation, UI mockup generation
• Media — Video scripting, thumbnail creation, audio transcription
• Business — Lead scraping, product listing, market analysis, pricing strategy
• Data — Cleaning, transformation, visualization, statistical analysis

Skill Anatomy

A skill file follows a consistent structure:

# Skill: Deep Research

## Purpose
Conduct thorough research on a topic with source
synthesis and citation tracking.

## Inputs
- topic: The subject to research
- depth: shallow | medium | deep
- format: report | bullets | wiki

## Steps
1. Identify 5-10 authoritative sources
2. Extract key claims with citations
3. Cross-reference conflicting information
4. Synthesize findings into requested format
5. Write to vault/outputs/

## Output
Structured research document saved to the Memory Vault.

## Guardrails
- Always cite sources
- Flag low-confidence claims
- Escalate if topic requires specialized knowledge

Pipelines

Pipelines are declarative YAML files in .claude/pipelines/ that chain multiple skills into automated workflows. The Orchestrator reads the pipeline definition and executes each step in sequence, passing outputs between stages.

Pipeline Structure

name: content-pipeline
description: End-to-end blog content creation
trigger: manual

steps:
  - skill: deep-research
    agent: researcher
    inputs:
      topic: "{{topic}}"
      depth: deep

  - skill: blog-writing
    agent: writer
    inputs:
      research: "{{steps.0.output}}"
      tone: professional
      length: 1500

  - skill: seo-optimization
    agent: marketing-hub
    inputs:
      content: "{{steps.1.output}}"
      keywords: "{{keywords}}"

  - skill: social-distribution
    agent: marketing-hub
    inputs:
      content: "{{steps.2.output}}"
      platforms:
        - twitter
        - linkedin

Pipeline Features

• Variable interpolation — Use {{variable}} syntax to pass data between steps
• Step references — Access previous step outputs via {{steps.N.output}}
• Conditional execution — Steps can have condition fields for branching logic
• Error handling — Each step can define on_error: skip | retry | abort
• Triggers — Pipelines can be triggered manually, via CRON schedule, or by webhook

Continuous Loops

AI OS supports CRON-scheduled autonomous routines through the Automator and Routine Runner agents. These loops execute predefined pipelines on a schedule with built-in rate limiting to manage API costs.

name: daily-intel
schedule: "0 8 * * *"  # 8 AM daily
agent: routine-runner
pipeline: market-intel-pipeline
rate_limit:
  max_runs: 1
  window: 24h

Custom Skills

You can create custom skills by adding Markdown files to .claude/skills/. Follow the standard anatomy (Purpose, Inputs, Steps, Output, Guardrails) and the system will automatically discover and make them available to agents.