Skip to main content

​
Skills & Playbooks

Skills are curated operational playbooks that teach agents how to handle specific workflows. They define procedures, routing rules, tool contracts, and output formats — giving agents domain expertise beyond their base system prompt. Skills are not tools. Tools execute actions; skills teach agents how to use their tools effectively.

​
Progressive disclosure

Skills use a three-level disclosure model to keep system prompts lean:
LevelWhenWhat the agent sees
Level 1Always (system prompt)Skill name + description — just enough for the agent to know the skill exists
Level 2On activationFull playbook body — detailed procedures, routing rules, output contracts
Level 3On demandSupport files (reference tables, data shapes, examples) within the skill directory
This means agents don’t pay the token cost of every playbook on every request — they load the full playbook only when the user’s request matches a skill’s description.

​
Built-in skills

SkillAgentWhat it teaches
customer-communications-playbookMainAgentCross-agent coordination for customer outreach — when to delegate to RecordsAgent, EmailsAgent, or MeetingsAgent
meetings-agent-playbookMeetingsAgentMeeting research, prep, transcript analysis, invite coverage, scheduling boundaries
emails-agent-playbookEmailsAgentEmail search, drafting, reply handling, thread context
records-agent-playbookRecordsAgentCRM queries, record resolution, warehouse data shapes
research-agent-playbookResearchAgentWeb search, company news, SEC filings, research workflows
tasks-agent-playbookTasksAgentTask creation, scheduled tasks, follow-up management
email-compositionCross-agentEmail drafting patterns and output formatting
email-reply-draftingCross-agentReply-specific drafting rules
application-contextMainAgentProduct vocabulary, navigation areas, CSM workflow terminology

​
Discovering skills

​
REST API

The current public Python SDK does not expose a dedicated client.skills resource. Use the REST endpoints directly for skill discovery. 
# List all skills
GET /api/v1/skills

# Get a skill by name (includes full playbook body)
GET /api/v1/skills/{name}

# Get skill content (main playbook or support file)
GET /api/v1/skills/{name}/content?file_path=references/data-shapes.md

# List skills for a specific agent
GET /api/v1/agents/{id}/skills

​
Skill structure

Skills are defined as directories containing a SKILL.md file with YAML frontmatter: 
skills/
  meetings-agent-playbook/
    SKILL.md                          # Main playbook
    references/
      warehouse-response-shapes.md    # Support file
  customer-communications-playbook/
    SKILL.md

​
SKILL.md format

---
name: meetings-agent-playbook
description: Use this skill for customer meeting support in CSM workflows. Activate whenever the task involves checking recent or upcoming meetings, validating invite coverage, proposing open time blocks, or using meeting transcripts for prep and follow-up.
agents:
  - MeetingsAgent
---

# Meetings Agent Playbook

Your goal is to do the legwork for customer meeting research and prep...

## Mission
...

## Working Boundaries
...

## Tool Contracts
...

​
Frontmatter fields

FieldRequiredDescription
nameYesUnique identifier (lowercase, hyphens, must match directory name)
descriptionYesWhen to activate this skill — the agent matches user requests against this
agentsNoList of agent names this skill applies to (empty = all agents)
metadataNoArbitrary key-value metadata

​
How agents use skills

  1. At startup, all skills are discovered from SKILL.md files
  2. The available_skills prompt block injects a compact catalog (name + description) into each agent’s system prompt
  3. When a user request matches a skill’s description, the agent activates it via the load_skill tool
  4. The full playbook body is loaded and the agent follows its procedures
  5. If the playbook references support files, the agent can load them on demand
This means the agent’s base prompt stays small (just the skill catalog), but the full operational knowledge is available when needed.