MIT Licensed · Open Source

Stop Re-Teaching Your AI Agents

A.S.K. is a versioned skill library for AI agents. Define a capability once. Invoke it forever. Skills compound. Knowledge stops evaporating.

Get the Quick-Start Bundle — $47 View on GitHub
~40%
Agent success rate before
 
>90%
Agent success rate after
The Problem

Every agent session starts from zero

Your deploy agent re-discovers your CI config. Your notification agent re-figures-out Telegram. Your purchasing agent re-learns your spending limits. Every. Single. Session.

You've built the same capabilities ten times. They work. Then the session ends and everything evaporates. The next agent starts from scratch.

"We spent more time re-teaching agents what they already knew than building anything new."

— Built from running 9 specialized agents in production at Foundry Familiars
The Architecture

Three tiers of reusable agent skills

Each skill is a versioned Markdown file with trigger conditions, input schema, methodology with reasoning, and output contract. Agents invoke skills by name. Skills compose into workflows.

Foundation

Portable primitives that work anywhere. Push to GitHub, send a Telegram message, fetch a URL.

ASK: foundation/github-push

Domain

Business-specific compositions. Full deploy cycles, purchase workflows, compliance checks.

ASK: gfs/tes-deploy

Orchestration

Multi-skill workflows with error propagation, retry logic, and fallback chains.

ASK: orchestration/release-train
How It Works

Skills are contracts, not functions

The SKILL.md file documents the why, not just the what. Future agents understand and adapt the methodology, not just execute it blindly.

# foundation/github-push/SKILL.md

description: "Push changes to a GitHub repository"
trigger: "When code changes need to be committed and pushed"

inputs:
  repo: string    # Target repository
  branch: string  # Target branch
  message: string # Commit message

methodology:
  # 1. Stage changes (why: prevents committing unintended files)
  # 2. Validate diff (why: catches accidental secret exposure)
  # 3. Commit with conventional message format
  # 4. Push with force-with-lease (why: prevents overwriting remote changes)

outputs:
  commit_sha: string
  push_status: success | conflict | error
Production Results

From running 9 agents in production

~40%
>90%
First-attempt success rate on complex tasks
Every session
Once
Times a capability needs to be defined
Hours
Minutes
Time to add a new agent capability
$47

Quick-Start Bundle

Everything you need to implement A.S.K. in your agent system this week. Built from a year of running agents in production.

27-page implementation guide
Python executor runtime
5 production-ready skills
Composition patterns
Error handling templates
Buy the Quick-Start Bundle

Instant delivery · Stripe-secured payment

The framework is free. Forever.

A.S.K. is MIT licensed. The Quick-Start Bundle is for teams who want the structured walkthrough, production patterns, and Python runtime instead of building from the spec.

Star on GitHub →
🎙️ VIDEO SERIES

The ASK Podcast

5 episodes deep-diving the Agent Skills Kernel architecture. Watch the full series free, then grab the bundle.

Episode 1

Intro & The Problem

Episode 2

What A.S.K Is

Episode 3

Three-Tier Architecture

Episode 4

No Vibe Coding & Dark Factory

Episode 5

What This Enables