Build once, run on any agent: a practical guide to agent skills

A developer's desk with a laptop, sticky-note notebook, code papers, phone, two nameplates and stickers. Demos What this is Slides The code GitHub September updates Tunes Andrea on X Shishir on LinkedIn Procore Survey
×

Demos

Proof Agent

Andrea's independent verifier follows a pull request through findings, corrections, and verdicts backed by evidence.

Get the skill and action

Dataforge

Shishir's four skills turn a project manifest and an issue into a data pipeline.

Get the skills and walkthrough
Also: proof-review, for verification under load
×

Build once, run on any agent

Most developers use AI the same way: open a chat, type a prompt, move on. The context disappears. The next step starts from scratch. Agent skills fix that. Andrea Griffiths builds them for her own recurring work: files any coding agent can run. Shishir Tewari builds them for a data engineering team at Procore, where a skill has to encode senior-engineer judgment and survive production. Different scale, same format. The file you write for yourself tonight is the same file that runs a team’s pipeline.

You will leave with a clear model for when to build a skill instead of writing another prompt.

Session summary

Read the eight-page AI-generated summary by VOXO from GitHub Universe 2026: reusable skills, evidence-based verification, and config-driven data pipelines.

demos/
Demo code and walkthroughs.
slides/
Presentation materials.
assets/
Images, diagrams, and recordings.