docs: move agent-exploration to wizard README, trim comments to current behavior

- README: add "Explore with an agent" under Running locally → Testing (was wrongly
  placed in the workbench README).
- scripts/README: drop the cross-PR pointer to the #703 repro scripts.
- Trim header/inline comments across the harness + scripts to concise descriptions
  of what the code does now — no history, no change-rationale.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

Author: gewenyu99

README.md

@@ -91,9 +91,7 @@ program e2e-drivable, add its profile to `profiles.ts`.
 
 The flow is **snapshot-tested** offline (no agent, deterministic):
 `__tests__/e2e-flow-snapshot.test.ts` golden-checks the (screen → decision)
-trace. Update with `jest -u` after an intentional flow/profile change. This is
-the structured-state analog of the TUI ANSI screenshots in
-`scripts/__screenshots__/`.
+trace. Update with `jest -u` after an intentional flow/profile change.
 
 ## Record & replay
 

e2e-harness/ARCHITECTURE.md

@@ -1,13 +1,10 @@
 /**
- * WizardE2eProfile — a program's declarative e2e "test definition": the
- * UI choices a headless e2e run should make at each decision point.
+ * WizardE2eProfile — a program's declarative e2e "test definition": the UI
+ * choices a headless e2e run makes at each decision point.
  *
- * The *choices* are product knowledge about a program's flow, but they live in
- * the harness ({@link ./profiles}, keyed by program id) rather than on the
- * program config — so none of this e2e machinery reaches the wizard's
- * production source. The harness is generic: it reads a profile and asks
- * {@link decideE2eAction} what to commit on the current screen. Add a program's
- * profile to {@link ./profiles} to make it e2e-drivable.
+ * Per-program choices live in {@link ./profiles}, keyed by program id.
+ * {@link decideE2eAction} maps the current screen + a profile to the commit to
+ * make. Add a program's profile to {@link ./profiles} to make it e2e-drivable.
  */
 
 import { ScreenId, Overlay, type ScreenName } from '@ui/tui/router';

e2e-harness/e2e-profile.ts

@@ -2,12 +2,10 @@
  * Per-program e2e profiles — the UI choices a headless run makes driving each
  * program's flow.
  *
- * Each program declares its test path as a readable JSON next to the program
+ * Each program declares its test path as JSON next to it
  * (`src/lib/programs/<program>/test/e2e.json`): a `profile` (the options the run
- * auto-takes) plus a documented `path`. We load the `profile` here and map it by
- * program id. Those JSONs are *data*, imported only by this harness — never by
- * prod code — so they don't reach the wizard's production source or its bundle.
- * Look one up with {@link profileFor}.
+ * auto-takes) plus a documented `path`. {@link profileFor} loads the `profile`
+ * and maps it by program id.
  */
 
 import { Program, type ProgramId } from '@lib/programs/program-registry';

e2e-harness/profiles.ts

@@ -26,9 +26,3 @@ none of it ships in prod. `WizardCiDriver` (read/act over the store), the
 screen→action registry, the `wizard-ci-tools` MCP server, the e2e profiles, and
 the recorder/replay. See [`ARCHITECTURE.md`](../e2e-harness/ARCHITECTURE.md) for
 how an agent drives these (env strip, scoped project id, gotchas).
-
-> **Security-leak repro scripts** (`relay-prod.no-jest.ts`,
-> `precedence.no-jest.ts`) that reproduce the `ANTHROPIC_BASE_URL`
-> settings-override gateway leak live on the fix PR
-> ([PostHog/wizard#703](https://github.com/PostHog/wizard/pull/703)), documented
-> in its description + comments.

scripts/README.md

@@ -1,13 +1,7 @@
 /**
  * Render a recording to per-frame TUI snapshots — one `.ans` file per key
- * moment, the REAL Ink screen rendered to ANSI (via replay's renderFrame, which
- * needs real ink, hence tsx not jest).
- *
- * These are the snapshots the workbench's visual-comparison flow diffs against a
- * committed baseline. A recording comes from a real `--e2e` run, so the
- * snapshots are what the user actually saw; run-to-run differences (e.g. the
- * agent enqueuing a different task) show up in the side-by-side for a human to
- * review.
+ * moment, the real Ink screen rendered to ANSI via replay's renderFrame (needs
+ * real ink, hence tsx not jest). Feeds the workbench visual-comparison flow.
  *
  *   tsx scripts/render-snapshots.no-jest.ts <recording.json> <outDir>
  */