AXM / public interoperability testAll tools

Keep the browser.
Re-check the procedure.

One disposable task. Four versions of the same screen. Can your caller reuse a known procedure, stop when its meaning changes, and catch a save that only looks successful?

Open the fixture Inspect the pinned contract

For an existing Hronaut-connected agent. No new integration or shared account.

Your browser, your workspace.

Create one scratch workspace in your own Hronaut instance. The test never connects to someone else’s browser, asks for credentials or sends the receipt anywhere.

The page keeps synthetic draft records locally. Your caller owns the decision to act and the check of what was actually saved.

A small test with a visible failure.

A layout change should preserve a valid procedure. A changed action should stop it. A misleading success message should fail the result check.

Only the receipt needs to cross between systems, after you review it.

The four cases

Case	Expected result	Save attempts
Baseline	PASS: requested draft persisted	1
Reordered layout	PASS: semantics survived	1
Changed action meaning	UNKNOWN: stop before task edits	0
False success message	FAIL: saved quantity is wrong	1, no retry

Those are expected outcomes, not prefilled results. This is an inspectable conformance test, not a blind benchmark. A task FAIL is the correct outcome for the last case.

The copy-ready prompt

Paste this into your already-connected Hronaut agent.

Run this disposable procedure-reuse interoperability test using your already-connected Hronaut tools. Read the pinned contract first: https://github.com/BigBirdReturns/axm-tools/blob/f6c8a39105398e916f3ab0197258f9ce43f9cb2a/procedure-reuse/v1/PROCEDURE.md

Create a new scratch workspace named "AXM procedure reuse interop"; use only its returned workspaceId and one tab for this run. Open https://bigbirdreturns.github.io/axm-tools/procedure-reuse/v1/fixture.html and check its exact origin and path. No sign-in, copied browser profile, installation, token discovery, terminal/browser substitution or policy changes.

The task is to save synthetic request R-1042 for DEMO-A / KIT-01 with quantity 3 and status DRAFT. Follow the contract's fixed procedure across all four cases. Before editing and again before saving, freshly inspect identity, unique controls and the action's meaning; the allowed effect is draft.local.save. Changed meaning or ambiguity means UNKNOWN before further task actions. An admitted case gets exactly one save, followed by Read saved record and a fresh snapshot. Inspect the persisted JSON, not the success message. Contradictory saved state means FAIL; missing evidence means UNKNOWN. Never retry or repair a case to make it pass.

Seal each observed verdict using the recorder. After case 1, advance to case 2, reload the same tab once, and check continuity without replacing the workspace. After case 4 export the fixture receipt. Report the actual four verdicts and save-attempt counts, whether all calls used the same scratch workspace, whether reload continuity held, and any tool errors or repairs. The expected sequence is PASS, PASS, UNKNOWN with no task edits/save, FAIL after one save, but report deviations as observed. Do not fabricate a Hronaut generation field or claim this tests reconnect or production safety. Keep workspace IDs, capabilities, cookies and tokens out of the shared receipt. Leave the task workspace available for inspection. If Hronaut is unavailable, report NOT RUN and stop.

Save the prompt as text

Now stale the browser lifetime without changing the page meaning.

The first run changed semantics while the workspace stayed live. The mirror test keeps the procedure semantically valid, reloads the same tab after a native Hronaut continuity checkpoint, and asks whether Hronaut vetoes the write until its own lifetime boundary is reconciled.

Open the copy-ready next test Inspect the v2 protocol

Expected: semantics still MATCH; Hronaut continuity BLOCKED; write blocked; fixture untouched; reconcile normally; fresh semantic admission; one verified save; PASS.

What we have actually run

Reference checks passed: 30 pure-rule tests and all four expected browser outcomes, including reload continuity, saved-record read-back and receipt export. The run used Playwright 1.57.0 with Chromium 143.0.7499.4. Independent Hronaut-native run: maintainer-reported on Hronaut 2.4.28 as PASS, PASS, UNKNOWN with 0 saves, then FAIL after one save; reload continuity held and no repairs were reported.

Inspect the reference CI run · Read its result · Read the maintainer-reported native observation

What this result can support

The contract tests reuse admission and fresh saved-record checks across controlled changes, in one workspace. It also checks the fixture marker across a page reload. The marker is not Hronaut’s workspace generation; the caller must verify its own workspace identity.

The reference executor uses Playwright, not Hronaut. Yevhen Tienkaiev subsequently reported the same four outcomes natively in one Hronaut 2.4.28 scratch workspace/tab. That is attributed maintainer evidence, not an AXM-operated run. The v2 mirror test now targets Hronaut's lifetime guard while page semantics remain unchanged. This test does not establish reconnect behavior, dispatch-time race protection, malicious-client resistance or production safety.

Publication PR · Versioned SHA-256 manifest · Reference browser checks · Maintenance and evidence boundaries

Capsule source: f6c8a39105398e916f3ab0197258f9ce43f9cb2a

Keep the browser.Re-check the procedure.