Ep 697 Tool 2:09 w/ Pippa & Tyler

Skillware AI Agent Skill Framework

Skillware is a Python framework that lets you equip agents with deterministic, modular skills, cutting out raw tool‑call boilerplate and letting you swap brains without touching the skill logic.

Embed this episode

Paste this on any site — the player is a self-contained iframe with no cookies or trackers.

<iframe src="https://sandrise.io/exploring-next/embed/697"
  width="100%" height="180" style="max-width:640px;border:0;border-radius:12px;overflow:hidden"
  title="Exploring Next — Episode 697 audio player"
  loading="lazy" allow="autoplay" referrerpolicy="strict-origin-when-cross-origin"></iframe>
Embed & API docs →
Script GPT-OSS 20B Voice Inworld TTS 2

Transcript

Pippa So, Tyler, I just hit up Skillware. It’s like…

Tyler What’s the deal?

Pippa You know, the whole out‑of‑the‑box tool‑call mess. Skillware is a Python framework that lets you "equip" an agent with a skill instead of writing the call yourself.

Tyler Right.

Tyler Target users…

Pippa Engineers who are building custom agents. They can pip install a skill, drop it into any LLM workflow, and swap brains without touching the skill logic.

Tyler Mm-hm.

Tyler Architecturally how does it stay deterministic?

Pippa Each skill is a bundle: logic, cognition, governance, interface. The logic is pure Python code, the cognition is a set of instructions, governance sets safety boundaries, and the interface standardises how the LLM calls it. That bundle survives across models.

Tyler Exactly.

Tyler No adapter rewrites? So you can just drop it into Gemini, Claude, GPT…

Pippa Yep. The SDK turns the bundle into a Gemini‑style tool with a system instruction. You just pass the tool in the generation config and the agent uses it. No extra glue.

Tyler Nice.

Tyler What about the actual skills? Any real‑world ones?

Pippa They’ve got wallet screening, a PII masker, and an issue resolver that can read a GitHub repo and spit out a ranked plan before you write any code. Each lives in its own pip extra: pip install skillware , , .

Tyler Cool.

Tyler Sounds promising, but deterministic execution can still bite if the skill logic is buggy. How do they guard against that?

Pippa They ship the skills pre‑compiled and come with tests in the repo. The registry lets you browse them before you inject, so you’re not reinventing tool definitions.

Tyler Got it.

Tyler Anything to try out right now?

Pippa Check out the GitHub repo: arpa hls/skillware. Just pip install, pick a skill identifier, and run the helper example from the docs. The docs walk through attaching it to Gemini or Claude.

Tyler Sounds doable.

Tyler All right, I’ll poke it in tomorrow. Maybe we’ll catch a bug if it tries to call the wrong API.

Pippa You always get the nitty‑gritty. We’ll hear about that later.

Tyler We’ll keep an eye on the registry. Until next time, keep those skills sharp.