skill tutorial · scenario
Measure learning outcomes with your agent
"Did the training work?" is really three questions you can collect — reaction, learning, behaviour — and one you usually cannot. Your agent can assemble the Kirkpatrick evidence stack across the feedback-form folders and the assessment presets. Here is the loop, including where it honestly stops.
The four levels, and what an agent can collect
Kirkpatrick's model (Kirkpatrick, Evaluating Training Programs, 1998) splits evaluation into four levels. The first three are collection problems an agent is good at: Level 1 reaction (a post-training survey), Level 2 learning (a pre/post delta from an assessment.json), Level 3 behaviour (a day-30 follow-up asking what changed). Level 4 results — business impact — is not a survey, and no folder pretends otherwise. The training-effectiveness guide and the Kirkpatrick model guide carry the full reasoning, including that boundary.
Hand it the evaluation you actually need
Say which levels and on what program, because the folders differ:
the brief
> I ran a 2-day training for 20 people. I need Level 1 reaction, > Level 2 learning (a pre/post on one measurable activity), and a > Level 3 day-30 behaviour check. Assemble the forms and the schedule.
Let it pull the right folders and presets
The agent maps each level to a folder that already ships it:
evaluation assets
curl -s https://workshopskill.com/en/skills.json | jq -r ' .skills[] | select(.name|test("survey|evaluation|feedback")) | [.name, .category, .duration_minutes] | @tsv' # Level 1 -> post-training-survey / training-evaluation-form # Level 2 -> the assessment.json inside any assessment-type skill # Level 3 -> the day-30 follow-up block in training-evaluation-form
Collect, and compile the delta
Once installed, the agent runs the collection. Reaction and the behaviour check are surveys; learning is the two-pass measure. On paper that is anonymous and needs no account; to generate the delta report online it uses formlm-cli, which is the one step with a human login. Then it compiles team-level deltas and the open answers into the one-page report a manager can forward. The pipeline that turns an assessment.json into that delta report is the delta-report walkthrough — this page is the wider stack that wraps it.
keep going
questions
What can an agent actually measure about a workshop?
It can assemble and run the collection for Kirkpatrick Levels 1 to 3 — reaction (did the room find it useful), learning (the pre/post delta from an assessment.json), and behaviour (a day-30 follow-up asking what actually changed). It builds the forms, schedules the re-check, and compiles the numbers. Level 4 — business results — is not something a folder measures; claiming a survey proves a profit line is exactly the mistake the Kirkpatrick guide warns against, and the agent will not fabricate it for you.
How is this different from the delta-report walkthrough?
The delta-report guide is one rung of this ladder: it takes a single assessment-type skill's pre and post measure and produces that pair-level delta. This page is the whole evidence stack around a training program — reaction survey plus the delta plus the delayed behaviour check — wired across the feedback-form folders and the assessment presets. Pick the delta report if you just want one activity's before/after; pick this if you need the L1-L3 story for a program or a client.
Do these measures prove learning happened?
They show a shift, honestly and at the team level; they are not psychometric proof. Every measure here is a two-minute field instrument, reported as the room's own before/after delta, never as a ranking of people. That is a real and useful signal and it is also a limit — the folders say so on the sheet, and the agent should keep that wording, not upgrade it into a claim the instrument cannot support.