Install the CLI
npm install -g @workbench-ai/workbench
Install the Workbench skill
npx skills add workbench-ai/workbench
Send this to your agent
Scan my local agent skills with Workbench
Ingest any work evidence as Sources. Keep the workflows that matter,then explicitly apply Evals before you improve Skills.
Evidence to durable behavior
Models organize the evidence. You decide what matters. Workbench preserves that judgment from Sources through Evals to Skills.
Ingest immutable records from any source. Analysis identifies recurring work and preserves exact evidence citations.
Explore as a hierarchy or optional 2D Map.
Inspect the workflow and its citations. Human judgment decides what is useful; nothing advances by default.
Decision recorded with exact citations.
You select kept workflows and confirm the objective. The model proposes an immutable draft; Apply creates the Eval, while Discard changes nothing.
Run the Eval when you choose, then use its proof to author and improve portable, versioned Skills.
Versioned instructions, references, and proof from the Eval you chose to run.
Nothing is applied or run automatically. Every transition remains inspectable.
Install the CLI
npm install -g @workbench-ai/workbench
Install the Workbench skill
npx skills add workbench-ai/workbench
Send this to your agent
Scan my local agent skills with Workbench
Rolling skills out across an organization? Talk to us about governance.