Interactive learning / local only

The practice studio

Make a decision, observe the consequence, and try again. These local exercises connect the learning guides to a workflow you can inspect before running a real lab.

Choose a simulation below: step through a bounded workflow, select relevant context, or run a small JavaScript fixture with a failing and a corrected implementation.

The workflow is an illustration, not a live agent. Context units are synthetic, not model tokens. Fixture checks execute only the fixed example in your browser; they do not establish that a course lab was completed.

01 / Ask → Plan → Agent

Agent workflow

Walk through a bounded change and see why a failed review must stop the handoff.

01AskPending
02PlanPending
03AgentPending
04ReviewPending
05EvidencePending

Numbers show the handoff order. The outlined stage is current; visited stages are labeled. Playback is opt-in and pauses when this page is hidden.

Ask

Investigate the repository, constraints and unknowns before requesting a change.

Expected artifact

A source-grounded map and open questions.

Plan

Define a bounded change, its risks, acceptance criteria and verification.

Expected artifact

A reviewable plan with explicit evidence gates.

Agent

Implement the approved scope, inspect tool results and iterate within the agreed limits.

Expected artifact

A focused diff and the checks actually executed.

Review

Challenge the implementation from a clean or specialized context.

Expected artifact

Findings tied to the acceptance criteria and concrete code.

Evidence

Compare the result with the original requirement. Record failures and unverified assumptions.

Expected artifact

Reproducible commands, outcomes and remaining limitations.

02 / Context workbench

Context workbench

Choose just enough relevant context. More files are not automatically better context.

Normalize names without changing unrelated behavior.

Include the task, implementation and tests. Stay within the illustrative budget; exclude unrelated material.

Task + acceptance criteriaTrim whitespace, normalize case and reject an empty name.2ImplementationThe function to change and its immediate callers.3Relevant testsExamples and boundary cases that describe the behavior.2Unrelated archived logsA large log from a different task.4Unrelated screen stylesCSS for a screen unaffected by name normalization.1

These units are invented for this exercise. They are not tokens, model limits or measurements of AI quality.

03 / Verification loop

Verification loop

Run a tiny JavaScript fixture in this browser. Compare a failing implementation with a corrected one.

Fixture source / JavaScript

function normalizeName(value) {
  if (!value.trim()) throw new RangeError("Empty name");
  return value.trim();
}

Verification loop
Input	Expected	Actual	Result
`" Ada "`	`ada`	—	Not run
`"LIN"`	`lin`	—	Not run
`" "`	`RangeError`	—	Not run

Choose an implementation, then run the three checks.

Only the fixed example shown here executes. No input is evaluated as code, no shell command runs, and these results are not saved as course completion.