Skip to main content

Workflow Skills

A workflow skill is reusable operational memory for a multi-step task. It describes the exact sequence of PUDA commands, verification steps, and failure handling an agent should follow to complete a task reliably.

Workflow skills sit above machine skills. A machine skill teaches an agent how to operate one machine safely. A workflow skill teaches an agent how to combine commands—and sometimes several machines—into a repeatable procedure.

When to create a workflow skill​

Create a workflow skill when:

  • You run the same procedure more than once, such as dispensing into a well, calibrating a device, or preparing a sample.
  • The task spans several machines, such as liquid handling followed by imaging or measurement.
  • The sequence has ordering constraints, verification gates, or cleanup steps that must not be skipped.
  • You want the agent to stop and report instead of improvising after a failure.

Start from machine skills​

Before writing a workflow skill, make sure machine skills exist for every machine the workflow touches. The workflow should reference those constraints instead of redefining them.

Ask the agent to inspect the machines involved:

List the machines required for this workflow and confirm that machine skills exist for each one.

For any machine without a skill, summarize the missing safety rules, limits, startup steps, and cleanup steps before we continue.

Write the workflow skill​

Describe the task in plain language first, then ask the agent to turn it into a skill:

Create a workflow skill for dispensing reagent A into well B3.

The workflow must:
- Home the robot if it has not already been homed.
- Pick up a clean tip.
- Move to the reagent A reservoir using the known safe approach path.
- Aspirate 50 uL.
- Move to well B3 using the known safe approach path.
- Dispense 50 uL.
- Verify the final position.
- Drop the tip.

A good workflow skill should include:

  • Purpose — what the workflow accomplishes and when to use it.
  • Prerequisites — required machine state, consumables, calibration, or prior steps.
  • Command sequence — exact PUDA commands in order, with parameters.
  • Verification gates — checks after important steps, such as position, pressure, or image capture.
  • Cleanup — steps that must run even when the workflow fails partway through.
  • Failure handling — optional instructions for what to do when a command fails, including when to stop and report.
  • Example runs — one or more complete command sequences the agent can follow.

Add optional failure handling​

Workflow skills can describe both the normal path and the safe stop path. If a command fails, the agent can ask the user for approval before retrying. If you do not want the agent to retry, say so explicitly in the workflow skill.

Add failure handling to the workflow skill:

- If any move command fails, stop immediately and report the target position, actual position, and command response.
- If aspiration or dispense fails, ask before retrying. If retries are not allowed, report the failure and leave the robot in its current state.
- If verification fails, stop the workflow and do not continue to the next step.
- Always run cleanup steps when it is safe to do so, even after a failure.

Test with a dry run​

Before running on real samples, ask the agent to plan the workflow without executing physical actions:

Using the workflow skill, plan a dry run for <workflow_name>.

List every command in order, the expected machine state after each step, and every verification gate. Do not run any commands yet.

Then run a small real test with conservative parameters. Confirm that startup, verification, cleanup, and failure handling all behave as expected.

Refine after real runs​

Update the skill with observations from each run:

Update the <workflow_name> workflow skill with these observations:

- The safe approach path to well B3 for first machine requires raising Z to at least 50 mm before any XY move.
- The workflow completes reliably with 45 uL instead of 50 uL for this reagent viscosity.

Keep the skill precise. The goal is for the agent to follow the same reliable procedure every time: check prerequisites, execute in order, verify at the gates, clean up at the end, and stop safely when something goes wrong.

References​