# Prompts used in the lesson slides

These are repeatable teaching inputs. Actual assistant responses vary. Prepared candidate code must be labelled as a teaching example rather than a captured model response.

## Episode 1, stage A: task clarification

```text
Read task.md and save-gate.before.ts. Explain the triggering sequence,
the expected behavior and the scope of the smallest change.
List any missing information before proposing code. Do not edit yet.
```

## Episode 1, stage B: verified context

```text
Read project-context.md, task.md, SaveAction.example.tsx and save-gate.test.mjs.
Summarize the existing UI contract, where the gate must live,
the available check command and what that command does not verify.
Do not infer a package manager or a runnable native app from these snippets.
```

## Episode 2: bounded implementation

```text
Use the incomplete gate as the starting point. Implement one retained gate
that rejects overlapping admission with the result "ignored" and releases
on success or failure. Preserve the task's rejected error for the caller.
Change only the gate and tests unless you identify a necessary caller change.
Explain that change before expanding scope. Do not add dependencies.
```

## Episode 3: verification

```text
Compare the gate and its caller with every acceptance case in task.md.
Run the supplied gate tests if execution is available. Report the exact
command and result. Explain which React Native behavior still needs a host
app/device check. Distinguish test results from code review conclusions.
```

## Episode 4, stage A: skill authoring

```text
Draft a review-save-flow skill from the review we just performed.
Include its trigger, required inputs, ownership and failure-path checks,
evidence-based finding format and review limits. Keep it report-only.
Do not claim that button props alone prove request behavior.
```

## Episode 4, stage B: a fresh evaluation

```text
Use the review-save-flow procedure to review evaluation-cases.md.
For this evaluation, do not open the answer key in review-skill/cases.md.
Classify supported issues and inspected non-issues, explain the triggering
sequence, and separate UI-feedback concerns from duplicate-request claims.
Do not edit the examples. Do not claim you ran React Native.
```

For a stronger evaluation, place only the skill procedure and the fresh cases in a clean workspace, without the answer key. Explicitly loading a procedure demonstrates its review logic; it does not test automatic skill discovery. Record discovery as a separate, tool-specific check if taught later.