From Tribal Knowledge to Team Skills: Structured Knowledge for Your AI Agent

Every team has institutional knowledge that lives in people’s heads. The deployment checklist that only Sarah knows. The data taxonomy that lives in a Google Doc nobody can find. The incident response procedure that changes every quarter.

RAG systems try to solve this by chunking documents into small pieces and retrieving the most relevant fragments. That works for factual lookups — but it falls apart when the AI needs complete, structured instructions to perform a task correctly.

Today we’re launching Skills — a new way to give your AI agent full, authoritative documents it can load on demand.

Skills vs. RAG: Different tools for different jobs

	RAG (Knowledge Bases)	Skills
Document size	Chunked into 500-token fragments	Loaded in full (up to 50KB)
Retrieval	Semantic similarity search	Explicit by name
Best for	Factual questions across many docs	Step-by-step procedures
Invocation	Automatic (context injection)	On demand (`/skill:name`)
Scope	Account or department	Account or department

Think of RAG as a search engine and Skills as a bookshelf. You search when you don’t know where to look. You grab a book when you know exactly which one you need.

How Skills work

Creating a Skill

A Skill is a markdown document with a name, title, department scope, and tags. Create one from the Skills page or via the API:

# Deployment Pipeline

## Prerequisites
- Ensure all tests pass on the `main` branch
- Check that staging environment is healthy

## Steps
1. Create a release branch: `git checkout -b release/v{version}`
2. Run the build: `npm run build`
3. Deploy to staging: `make deploy-staging`
4. Run smoke tests: `npm run e2e:smoke`
5. If smoke tests pass, deploy to production: `make deploy-prod`
6. Tag the release: `git tag v{version}`

## Rollback
If anything goes wrong after production deployment:
1. Revert to previous version: `make rollback`
2. Notify #engineering in Slack
3. Create an incident report

Using a Skill in conversation

Invoke a Skill by name in the chat:

You: /skill:deployment-pipeline Deploy version 2.4.1 to production

The agent loads the full deployment pipeline document into its context and follows it step by step. It doesn’t guess or improvise — it has the complete, authoritative instructions.

Agent-initiated Skill discovery

The conversational agent also has two built-in tools for proactive Skill use:

list_skills — discover available Skills for the current department
load_skill — load a Skill by name when the agent determines it needs specialized knowledge

This means you can ask “deploy the latest version” and the agent will:

Recognize this is a deployment task
Call list_skills to check available Skills
Find and load deployment-pipeline
Follow the documented procedure

Department scoping

Skills are scoped to departments, just like recipes and workflows. The Engineering team’s deployment playbook doesn’t clutter the Marketing team’s Skill list. An admin can create account-wide Skills that are visible to all departments.

Practical use cases

Engineering:

Deployment procedures per environment
Incident response runbooks
Code review checklists
Architecture decision records

Operations:

Standard operating procedures
Vendor onboarding checklists
Meeting facilitation guides

Sales:

Discovery call frameworks
Objection handling playbooks
Competitive positioning guides

Compliance:

Audit preparation checklists
Data classification procedures
Regulatory reporting workflows

Limits and caching

Skills are full documents injected into the conversation context, so we enforce sensible limits:

50KB max per Skill (~12,000 tokens)
100 Skills per account, 50 per department
3 Skills loaded simultaneously in one conversation turn
Skill names are cached in Redis (5-minute TTL) to avoid Firestore reads on every message

If loading a Skill would push the conversation over the model’s context limit, the agent warns you and suggests starting a new conversation or using a model with a larger context window.

Getting started

Navigate to Skills in the sidebar
Click Create Skill
Write your markdown content
Set department scope and tags
Use /skill:name in any conversation

Skills are available on Pro plans and above. Create your first Skill.