Setup
Overview
Find the path to create, evaluate, improve, and publish a Workbench skill.
Workbench mines reviewed workflows from Sources, turns them into Evals, and uses those Evals to improve and publish versioned Agent Skills.
Choose a path
| Goal | Start here |
|---|---|
| Ingest evidence and discover workflows | Sources |
| Build your first measured skill | Quickstart |
| Learn the eval source model | Evaluation basics |
| Author cases and grading | Cases and grading |
| Configure agents and models | Agents and models |
| Choose tools for Office, PDF, and other file artifacts | File formats |
| Improve from failed or below-perfect evidence | Improve from evidence |
| Review candidate changes | Review candidate changes |
| Rerun proof evidence | Rerun proof evals |
| Inspect results | Results |
| Follow runs and jobs | Runs and jobs |
| Inspect traces, artifacts, and files | Traces, artifacts, and files |
| Review versions and lineage | Versions and lineage |
| Publish a Skill package | Publish |
| Install or clone a skill | Install and clone |
| Manage visibility and hosted operations | Visibility and Cloud |
| Look up command behavior | CLI reference |
| Understand installable skill packages | Skill packages |
Core concepts
| Concept | Meaning |
|---|---|
| Source | An ingested evidence corpus with immutable snapshots, grounded workflow Analyses, and explicit human review. |
| Eval | The grading standard for a workflow: cases, grading config, agents, and runtime environment source under .workbench/. |
| Skill | A standards-compliant Agent Skill package with SKILL.md plus support files. |
Runs are Eval evidence; package versions and handles are Skill lifecycle records, not additional product roots.
Daily loop
Start with Quickstart, then use the pages above for the Source, Eval, or Skill work you need to finish.