Declarative skills
Author workflows as plain YAML + Liquid-templated Markdown. No code.
Declarative workflows for VS Code chat
Write branching YAML playbooks that orchestrate your language model and MCP tools — with review and input steps — and run them right from the @skiller chat participant.
Predictable automation, not an autonomous agent
Section titled “Predictable automation, not an autonomous agent”Skiller runs skills — declarative, multi-step workflows defined in YAML and Markdown that call a language model, invoke your MCP tools, and pause for your input exactly where you tell them to. It is deliberately not a free-form agent and not an “Agent Skills” auto-loader — you write the steps; the model fills them in.
Human-in-the-loop
confirmation steps pause for your choice and can branch or jump with goto.
MCP tool orchestration
Call your configured MCP tools from llm and tool steps.
Live execution graph
Watch a skill run as a live execution graph, with branches and loops lighting up as they fire.
Layered discovery
Workspace, user, and built-in skills — with workspace winning.
Typed & validated
Manifests are schema-checked (Zod) with helpful errors before they run.
Free-form agent vs. Skiller
Section titled “Free-form agent vs. Skiller”| Free-form agent | Skiller | |
|---|---|---|
| Control flow | Model decides what to do next | You define the steps; the model fills them in |
| Side effects | Can act on its own | Runs only the steps you wrote, pausing where you ask |
| Reproducibility | Varies run to run | Same playbook, same shape every time |
| Branching | Implicit | Explicit confirmation steps with goto jumps |
If you want predictable, reviewable automation that still leverages an LLM — not an autonomous agent — Skiller is for you.