Skip to content

refactor(maestro): load the engine as a managed replay plugin - #4505

Open
thymikee wants to merge 1 commit into
fix/c3-prep-atomic-layer-20261114from
t3/maestro-plugin-extraction
Open

thymikee wants to merge 1 commit into
fix/c3-prep-atomic-layer-20261114from
t3/maestro-plugin-extraction

Conversation

@thymikee

@thymikee thymikee commented Nov 13, 2026 •

Copy link
Copy Markdown
Member

Summary

Extract managed Maestro into its v2 optional plugin for #3377, retaining selected-package loading and explicit host routing.

Unpublished candidate: 79c2006edca27fd335a939d02b1aad38c0ed1ff8. 104 files / 2,472 gross lines above PREP; 273 / 7,420 against default-main merge-base.

Validation

Independent Codex recheck found no actionable findings within the R1 repair and preservation interactions. The two-file repair survives all four rebases; eight range-diff entries match.

At 79c2006edca27fd335a939d02b1aad38c0ed1ff8, pinned frozen install/build and full-repository format passed. Exact affected: 42 passed / 25 skipped, exactCommit=true, checkoutChanged=false; receipt 7063f85786a69c0c8b4fa923828f12b615def3d74982f9fc9e14bd794bea4f41. Full: 41 passed / 25 skipped; receipt 7e1f19f6ecb5c4551041126f11f738a438f914bbb5b9869e2a009edb85835e5e.

All seven new-head canonical invocations passed once, serially, under one frozen ENV. Counts overlap; skips retain their CI/manual owners. Existing refs stayed unchanged; source/logical index matched before and after each gate. Physical-index captures remain separate, with within-head byte-change causes unproven. Initial/final snapshots cannot exclude transient edits.

The candidate remains unpublished; published head 924a0bd3 is unchanged and new-head hosted checks are unverified. Node22.13 packaging and earlier ENV/ref owner dispositions still block source publication. Both plugins remain private pending maintainer placeholders, trusted publishers and usable publication. Native ADR, live/current-client, performance and security remain open. No merge clearance.

@github-actions

github-actions Bot commented Nov 13, 2026 •

Copy link
Copy Markdown
Contributor

Size Report

Metric Base Current Diff
Installed (including dependencies) 6.31 MB 6.08 MB -225.6 kB
Package (unpacked) 6.30 MB 6.08 MB -225.6 kB
Package (download) 1.91 MB 1.84 MB -67.8 kB

Startup median (7 runs, lower is better):

Scenario Base Current Diff
CLI --version 27.8 ms 23.1 ms -4.7 ms
CLI --help 59.0 ms 49.4 ms -9.6 ms

@thymikee

Copy link
Copy Markdown
Member Author

At bea660f, two likely defects need checking before this is ready for human review. There are no conflicts.

Could the bundled dependency configuration give the installed plugin a private diagnostics AsyncLocalStorage, causing its direct emitDiagnostic calls to return without recording anything? That would lose maestro_tap_dispatch and maestro_screenshot_compare events in debug and request logs, while source-symlink tests keep passing. Please pass diagnostics through ReplayBackendHost, as with registerSensitiveValue, and pin the host-state boundary with a package check. This conclusion comes from code inspection; it still needs a packed-artifact run.

With an empty plugin store, could startAgentDevice({ plugins: [maestroPlugin()] }) fail to replay a YAML flow because withReplayBackend opens the caller’s store rather than using the embedded host’s registry? A different installed version would also make source collection and execution use different engines. Please give the client the host’s resolver and add an imported-plugin, empty-store embedded test. This path was inspected, not executed.

The 15 failed checks look related: 13 mutation shards lack the plugin installation supplied by the unit-test global setup, and their missing reports fail the aggregate; Repo Guards fails because the compound check:package script is no longer credited by verbatimScripts. Please fix those setup and gate-model paths.

Could an explicit resolver on the client and writer hosts replace the ambient binding, per-flow registries and global test setup? The CLI could share one store scope per command, filtered to the declared package, while the embedded client uses its runtime’s resolver. Would adding an optional ReplayBackendResolver to those host contracts be enough to make that smaller design possible first?

Not blocking: count only active backends in shutdown reports, open Maestro lazily for its fuzz target, and keep the remaining test/type cleanups and #3871 failure-classification work separate.

After the fixes, please attach packed-dist Android and iOS replay results, including selector taps, an animation or screenshot wait, and debug output with both diagnostic events above; also include the ADR 0015 Android timing comparison and the empty-store embedded run. The placeholder publication and trusted-publishing setup remain separate draft blockers.

@thymikee

Copy link
Copy Markdown
Member Author

bea660f now conflicts with the target branch, 0.22. Please rebase onto 0.22 and address the outstanding review in the same push. I will re-check the updated head.

@thymikee
thymikee force-pushed the t3/maestro-plugin-extraction branch from 9e95c90 to 87c8830 Compare November 13, 2026 16:11
@thymikee

Copy link
Copy Markdown
Member Author

Thanks for the rebase. The two new commits fix the mutation shard and Repo Guards failures from bea660f, and all 32 checks pass at 2e004ce. Two problems from my earlier review (#4505 (comment)) are still in the code, and the live validation is still missing. There are no conflicts.

The bundled engine still loses its diagnostics. packages/maestro/tsdown.config.ts:19 bundles every @agent-device/* import, so the plugin carries its own copy of host-kit diagnostics with its own module-level AsyncLocalStorage (diagnostics.ts:61). The engine calls emitDiagnostic directly (daemon-runtime-tap.ts:235, maestro-screenshot-comparison.ts:74 and :122). In the installed dist that copy has no store, so it returns early and the maestro_tap_dispatch and maestro_screenshot_compare events never reach --debug or the request diagnostics. Those events are the dispatch evidence the ADR 0015 timing work uses. Workspace tests resolve the shared module, so CI cannot see this. The rule should be: the bundled engine reaches per-request or per-runtime host state only through ReplayBackendHost. Please add emitDiagnostic to the host the way registerSensitiveValue is already passed, and add a package-boundary rule that forbids packages/maestro/src from importing host-kit diagnostics, host-kit/request* and runtime slots.

Embedded hosts still split the plugin set. startAgentDevice accepts plugins (host-types.ts:14) but gives them only to the embedded daemon and builds the client with { host } (host.ts:73). The client's replay and test writers call loadReplayScriptSourceBundle outside the runtime slot, so withReplayBackend (replay-backend-registry.ts:40) finds no bound resolver and opens the caller's plugin store. With startAgentDevice({ stateDir, plugins: [maestroPlugin()] }) and an empty store, replaying a flow.yaml with maestro fails with plugin-not-installed. With a different store version, source collection and execution would use different engine versions. run-maestro-flows.md:21 tells users the opposite. The rule should be: source collection and export resolve through the same plugin set as the host that executes. Please pass the embedded runtime's resolver to its client host, let withReplayBackend prefer a host-supplied resolver over the store, and add an embedded-host test with an imported plugin and an empty store.

Before merge, please run the packed plugin dist, not the workspace symlink (scripts/install-workspace-maestro.ts with source=false, or pnpm pack and plugins add), through a daemon started from this head. On Android and iOS, replay a flow.yaml with --maestro that reaches selector taps and an animation or screenshot wait. Show the replay success output, and with --debug show the request diagnostics containing maestro_tap_dispatch and maestro_screenshot_compare. Add the ADR 0015 Android timing comparison against upstream Maestro. Also show the embedded startAgentDevice run with an imported maestroPlugin() and an empty store replaying a YAML flow. I did not run a packed-dist install; both findings come from reading the code, and I did not execute the embedded path. I also could not confirm from the check data that the 13 mutation shards ran rather than being path-skipped, and I did not re-review full replay, test and export parity in this pass.

Not blocking: the Validation section in the PR body still names bea660f, "86 files; 2,393 gross lines"; it could name 2e004ce on 4d75aa4 with 88 files and 1,685 gross lines, and say which review items this push addressed. Take it or leave it.

After the two fixes and the live run, package publication (private: true, trusted publishing) is the remaining draft gate.

@thymikee
thymikee changed the base branch from 0.22 to t3/maestro-export-tests-prep November 13, 2026 21:57
@thymikee
thymikee added this pull request to stack #4529 November 13, 2026 21:57
@thymikee
thymikee force-pushed the t3/maestro-plugin-extraction branch 2 times, most recently from 10eba62 to 8e6c1fe Compare November 13, 2026 22:12
Base automatically changed from t3/maestro-export-tests-prep to 0.22 November 13, 2026 22:29
@thymikee
thymikee force-pushed the t3/maestro-plugin-extraction branch from 118c264 to 284b35f Compare November 13, 2026 22:29
@thymikee

Copy link
Copy Markdown
Member Author

I reviewed 284b35f against the earlier findings at #4505 (comment). The code fixes landed: the engine now gets diagnostics and process execution through the host, and the packed embedded empty-store run is covered by packed-runtime 'embedded-worker'. CI is green at this head (25 checks, none failing), and no conflicts are known.

One item is still open. This change moves device-facing Maestro dispatch (selector taps, animation and screenshot waits, runScript children) into a bundled, separately installed plugin. CI drives the packed dist only with a fake invoke (packages/maestro/test/packed-runtime.mjs). No run has reached a real device through a daemon started from this head with the packed plugin installed in the store, and the PR body still says the run is underway. So nobody has seen real tap timing, the screenshot worker on real captures, or the ADR 0015 timing parity. Please install the packed dist (pnpm pack and plugins add of the tarball, or scripts/install-workspace-maestro.ts in source=false mode, not the workspace symlink) and start a daemon from 284b35f or a later head. On one Android and one iOS device, run replay flow.yaml --maestro --debug with a flow that reaches a selector tapOn and a waitForAnimationToEnd or screenshot wait. Paste the success output and the maestro_tap_dispatch and maestro_screenshot_compare diagnostics lines, and add the ADR 0015 Android timing comparison against upstream Maestro. After that, the private:true and trusted-publishing draft gate is the next step before merge.

Not blocking, take or leave: scripts/layering/architecture-ownership.ts:154 still has no rule that keeps packages/maestro/src off host-kit diagnostics, request, runtime-slots and command imports (exempting the child entries already in PROCESS_ENTRIES), so a future direct import would pass the workspace tests; src/host.ts:52 passes the replay resolver through an extra field on the composeDaemonRuntimeDeps return type that the daemon never reads, and a local variable captured in the composeDeps callback would avoid that; and DiagnosticEventInput and DiagnosticLevel are now spelled both through @agent-device/contracts/host-diagnostics and the host-kit re-export in packages/host-kit/src/internal/diagnostics.ts:13, so moving the 8 importers and dropping the re-export would leave one spelling.

@thymikee

Copy link
Copy Markdown
Member Author

The rebase to 0ef1318 keeps all 25 checks green, and the delta is one upstream conformance fixture line from #4538. The packed-plugin evidence asked for in the earlier review (#4505 (comment)) is still missing, so this is not ready to merge yet.

The PR moves device-facing Maestro dispatch (selector taps, animation and screenshot waits, runScript children) into a bundled plugin that users install separately. CI exercises the packed dist only with a fake invoke (packed-runtime.mjs:164). The PR body gives only a summary of a packed run at 284b35f, with no output or diagnostics lines. So nobody has shown real tap dispatch (daemon-runtime-tap.ts:235) or the screenshot worker on real captures from the installed tarball, and ADR 0015 Android timing parity is unchecked.

Please install the packed tarball, not the workspace symlink (pnpm pack and agent-device plugins add , or scripts/install-workspace-maestro.ts with source=false). Start a daemon from 0ef1318 or later. On one Android and one iOS device, run replay flow.yaml --maestro --debug on a flow with a selector tapOn and a waitForAnimationToEnd or screenshot wait. Paste the success output, the plugins list line that shows the store-installed version, and the maestro_tap_dispatch and maestro_screenshot_compare diagnostics lines. Add the ADR 0015 Android timing comparison against upstream Maestro.

After that, the private:true and trusted-publishing draft gate is the next step before merge.

@thymikee

Copy link
Copy Markdown
Member Author

Thanks for the update. The build config change at 2f016c0 looks fine, and all 24 checks pass, including the packed-runtime and package-smoke jobs. One thing is still missing before this can merge.

Device-facing Maestro dispatch (selector tap, animation and screenshot waits, runScript children) now runs from a separately installed packed plugin, but CI drives it only with a fake invoke (packed-runtime.mjs#L164). No run has gone through a daemon started from this head with the packed tarball in the plugin store and a real device. So real tap timing, the screenshot worker on real captures and ADR 0015 Android timing parity are unverified for the installed artifact. The issue and the earlier reviews ask for this before merge. Please pack the plugin and install the tarball, not the workspace symlink. Start a daemon from 2f016c0 or later. On one Android and one iOS device, run replay flow.yaml --maestro --debug on a flow that reaches a selector tapOn and a waitForAnimationToEnd or screenshot wait. Paste the success output, the plugins list line showing the store-installed version, and the maestro_tap_dispatch and maestro_screenshot_compare diagnostics lines. Please also paste the Android timing against the ADR 0015 baseline.

I did not run the build or inspect the packed d.mts, so I inferred the type-only effect from the import kind and the PR body. I could not check live device behavior. The branch has no conflicts.

@thymikee

Copy link
Copy Markdown
Member Author

Thanks for checking the public-type fix. The requested exact-head device/timing acceptance is still open; I am keeping this draft. Here is the concrete evidence already captured, with its actual source attribution rather than calling it a new 2f016 execution.

At source 0ef131824533ee97eed578a8f82c16f2e8485926, the real packed Maestro store reported:

{"backend":"maestro","source":"store","package":"@agent-device/maestro","version":"0.0.0","state":"active"}

Its managed factory was apiVersion 2 / replay-backend / maestro, without a symlink. Plugin tar SHA256 68c00ada2a2b9e4f292cff431a0bdb932690bc7abc5e7e27b89f6485385faf01; entry SHA256 163213f278f194f11e12a18b1fbda5e6e8ef4c39018eade28778045cfb2d2131. The T3-bound 0.21.31 client reached the source-built 0ef daemon (codeOrigin: checkout, codeSignature: graph:126:c5d88f294026f80f51201b848e527af805cafc13). The updated five-step optional-scroll scenario passed both engines on iOS and Android: all four exits 0; optional steps 3/4 warned, final Home passed. That scenario declares no engine invariants.

Earlier selector/animation paths were actually executed at 284b35f, not the new head. Retained raw diagnostic fields include:

{"phase":"maestro_tap_dispatch","data":{"kind":"selector","selectorKey":"text","expectedPoint":{"x":338,"y":822}}}
{"phase":"maestro_screenshot_compare","data":{"purpose":"animation","result":"stable","differencePercent":0.0006035887973621569}}

Those are inherited iOS trace excerpts, not fresh head performance or Android diagnostic attestation. The Android run also used the bundled android-helper 0.21.38-dev / API2; packed include/replay and authored screenshot execution succeeded there. Runtime-byte identity across the later declaration-only fix does not turn these into new executions.

ADR0015 timing is not admitted: unchanged full Pager correctness runs were base 6/9, extracted 6/9, upstream 7/9, with the exact failure locations retained; React Navigation ARM Release/Debug builds are blocked by the pinned Reanimated/Worklets output-path mismatch. I have not timed a passing subset or changed assertions. The Android shutdown route subsequently reported success with console KO while inventory remained booted; cleanup is explicitly partial and that route stopped.

Current 2f016 frozen install/build, affected 42/25 and full 41/25, plus hosted checks passed. These establish code/validation evidence separately from fresh current-head two-platform selector/animation execution, canonical timing and npm placeholder/trusted-publisher clearance. Local evidence roots: C3-report.txt under /private/tmp/c3-4505-0ef1318-20261114/ and /private/tmp/c3-4505-284b35f-3mq5bP/; exact receipts under /private/tmp/v022-six-exact-9n6ttv8l/.

@thymikee
thymikee marked this pull request as ready for review November 14, 2026 08:08

@cubic-dev-ai cubic-dev-ai Bot left a comment •

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

All reported issues were addressed across 105 files

Reply with feedback, questions, or to request a fix.

View guided diff | Turn on auto-fix | Re-trigger cubic

Comment thread .changeset/optional-maestro-engine.md Resolved
Comment thread scripts/fuzz/targets.ts Resolved
Comment thread website/docs/docs/run-maestro-flows.md Resolved
Comment thread scripts/check-affected/model.test.ts Resolved
Comment thread src/__tests__/replay-plugin-setup.ts Resolved
Comment thread .changeset/optional-maestro-engine.md Resolved
@blacksmith-sh

This comment has been minimized.

@apex-sentinel-staging

apex-sentinel-staging Bot commented Nov 14, 2026 •

Copy link
Copy Markdown

Apex Sentinel · ✅ Approve

Since the last review: 1 new · 0 still open · 0 no longer raised

1 finding(s): 1 minor.

🟡 Minor (1)

Where Finding
packages/replay-port/src/daemon-port/replay-backend-registry.ts:57 🆕 Client replay operations now evaluate every installed provider plugin package
Not in the reviewers' prompt (1 file)
  • pnpm-lock.yaml: generated or a lockfile, so its lines were left out

Reviewed with this repository's own Sentinel configuration (d78be6c88715).

💬 Talking to Sentinel

In a finding's thread, reply with:

Command What it does
@apex-sentinel-staging note <context> Adds context; Sentinel re-checks the cited code
@apex-sentinel-staging pivot <why> Says the finding is wrong; Sentinel re-checks the cited code
@apex-sentinel-staging like Marks the finding as useful
@apex-sentinel-staging dislike <why> Marks it as not useful, so it is not raised again

On this conversation, comment @apex-sentinel-staging review to run a full review once again.

Earlier reviews

Review of dda310eb · 14.11.2026 20:51 UTC

Apex Sentinel · ✅ Approve

Since the last review: 0 new · 0 still open · 1 no longer raised

No open findings.

Not in the reviewers' prompt (1 file)
  • pnpm-lock.yaml: generated or a lockfile, so its lines were left out

Reviewed with this repository's own Sentinel configuration (d78be6c88715).

Review of 0158fac9 · 14.11.2026 10:20 UTC

Apex Sentinel · 🟠 Request changes

Since the last review: 1 new · 0 still open · 0 no longer raised

1 finding(s): 1 major.

Required fixes

  • A transient replay engine load failure is cached for the daemon's whole life

🟠 Major (1)

Where Finding
packages/plugins/src/registry.ts:286 🆕 A transient replay engine load failure is cached for the daemon's whole life
Not in the reviewers' prompt (1 file)
  • pnpm-lock.yaml: generated or a lockfile, so its lines were left out

Reviewed with this repository's own Sentinel configuration (d38e35e61193).

Review of e0d48592 · 14.11.2026 10:07 UTC

Apex Sentinel · ✅ Approve

Since the last review: 0 new · 0 still open · 3 no longer raised

No open findings.

Not in the reviewers' prompt (1 file)
  • pnpm-lock.yaml: generated or a lockfile, so its lines were left out

Reviewed with this repository's own Sentinel configuration (d38e35e61193).

Review of 2f016c06 · 14.11.2026 08:16 UTC

Apex Sentinel · ✅ Approve

3 finding(s): 3 minor.

🟡 Minor (3)

Where Finding
packages/command-surface/src/replay/index.ts:293 test --maestro opens and disposes the plugin registry once per flow file
scripts/vitest-tmpdir-global-setup.ts:58 Global test setup installs the Maestro plugin for every suite in the run
src/host.ts:43 Daemon-side replay loses the bound resolver when composeDeps is overridden
Not in the reviewers' prompt (1 file)
  • pnpm-lock.yaml: generated or a lockfile, so its lines were left out

2 findings below minor hidden (user_settings.min_severity).

Reviewed with this repository's own Sentinel configuration (d38e35e61193).

@apex-sentinel-staging apex-sentinel-staging Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Apex Sentinel: 3 new findings on 2f016c0. The verdict and the full list are in the summary comment.

Comment thread scripts/vitest-tmpdir-global-setup.ts Resolved
Comment thread src/host.ts Resolved
@thymikee

Copy link
Copy Markdown
Member Author

The newly enabled Android smoke lane now supplies concrete head execution evidence, but it is red: the emulator booted and the fixture completed earlier scenarios, then smoke:maestro-clickable-first failed at the valid no-index tapOn replay with INVALID_ARGS, details.reason: plugin-not-installed, details.package: @agent-device/maestro. This is a missing migration prerequisite in the smoke workflow, not an emulator boot failure or evidence that the selector engine executed incorrectly. Failed job.

The workflow calls pnpm gate build and starts CLI/device tests without selecting the optional workspace plugin. I prepared the minimal fix: set the lane's isolated AGENT_DEVICE_HOME and invoke its existing built-workspace installer before daemon cleanup/start. The C4 layer must use its renamed existing installer for both plugins so the subsequent native suite also has its declared execution dependency. No scenario/assertion/timeout/selector fallback is changed. These patches are queued until the owned checkout's integration validation and the active head freezes release; they are not published fixes or passing receipts yet. I will validate the actual install/consumer path and push the corrected layers once, rather than rerun this unchanged failure. This PR remains ready for review with smoke execution still a merge blocker.

@thymikee

Copy link
Copy Markdown
Member Author

At 2f016c0 this branch now conflicts with its base 0.22. Please rebase it onto the current 0.22 and push, so CI and the stacked PRs (#4534, #4535, #4537) run against the right base. The earlier review still applies: the live packed-plugin replay runs on Android and iOS are the remaining merge condition.

@thymikee
thymikee force-pushed the t3/maestro-plugin-extraction branch from 2f016c0 to e0d4859 Compare November 14, 2026 09:49

@cubic-dev-ai cubic-dev-ai Bot left a comment •

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

All reported issues were addressed across 23 files (changes from recent commits).

Reply with feedback, questions, or to request a fix.

View guided diff | Turn on auto-fix | Re-trigger cubic

Comment thread .github/workflows/android.yml Resolved
Comment thread .github/workflows/apple.yml Resolved
Comment thread scripts/fuzz/execute.ts Resolved
Comment thread scripts/fuzz/harness.test.ts Resolved

@apex-sentinel-staging apex-sentinel-staging Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Apex Sentinel: 1 new finding on 0158fac. The verdict and the full list are in the summary comment.

Comment thread packages/plugins/src/registry.ts Resolved
@thymikee

Copy link
Copy Markdown
Member Author

Thanks for the update. I reviewed 0158fac against the earlier review (2f016c0 conflict). Most review threads are now fixed, but three problems remain, so this is not ready to merge yet.

The new job-level AGENT_DEVICE_HOME: ${{ runner.temp }}/... in android.yml:34 and apple.yml:39 is invalid. The runner context is not available in jobs.<id>.env, so GitHub rejects both files. Runs 43584109661 and 43584110724 at this head failed with zero jobs ("workflow file issue"). No Android or Apple smoke job runs on this PR, so the new Maestro install steps never execute, and the checks list hides this because the failed runs have no check runs. Please set the home at step scope instead, for example a first step that appends AGENT_DEVICE_HOME=$RUNNER_TEMP/agent-device-<lane>-smoke-home to $GITHUB_ENV. Then confirm both workflows start and their smoke jobs pass at the new head.

The private-package thread is only half answered. The changeset and run-maestro-flows.md now say the package is private, but the hint in replay-backend-declarations.ts:44 and website/docs/docs/plugins.md:61 still tell users to run agent-device plugins add @agent-device/maestro. packages/maestro/package.json is private at 0.0.0 and the changeset bumps only agent-device. A 0.22 CLI therefore removes Maestro replay and points replay --maestro, test --maestro and replay export users at an install that npm cannot resolve. Every surface that names the install command (the error hint, plugins.md, run-maestro-flows.md, the changeset) must name a package the shipping release can install. Please make @agent-device/maestro publishable in this PR: remove private, give it a real version, and add a changeset entry. If publishing is deliberately deferred, keep this PR off the 0.22 release until it happens, and do not pair a working-install hint with a "not yet available" doc.

The live-evidence ask is still open. packed-runtime.mjs:164 drives the packed plugin through a fake invoke only, and no run uses a daemon with the packed tarball in the plugin store on a real device. Selector tap dispatch, animation and screenshot waits on real captures, runScript children and the ADR 0015 Android timing are unverified for the installed artifact. Please install the tarball with plugins add (not the workspace symlink) and start a daemon from 0158fac or later. On one Android and one iOS device, run replay flow.yaml --maestro --debug on a flow that reaches a selector tapOn and a waitForAnimationToEnd or screenshot wait. Paste the success output, the plugins list --json line showing the store-installed version, and the maestro_tap_dispatch and maestro_screenshot_compare diagnostic lines. Also paste the Android timing against the ADR 0015 baseline.

Not blocking: the new it.each in harness.test.ts:253 creates a mkdtemp directory per case and never removes it, unlike its sibling tests, so a try/finally with fs.rmSync would fix it. You can take or leave this.

Simplicity question: could PluginInventoryEntry identity be a discriminated union like PluginCleanupOutcome, that is { kind: 'provider'; provider } | { kind: 'replay-backend'; backend } | { kind: 'unidentified' }? Lease admission would then match provider and unidentified variants by type, and the guard at lease.ts:441 would not be needed. The registry could also hand createPluginClose the list of cleanup owners it will dispose, instead of the daemon re-deriving it from inventory state. A future row-kind mix-up would then fail to compile. It needs a contract change in packages/contracts/src/plugin-inventory.ts, readPluginInventory and the daemon.json readers. This PR introduces the backend field, so it can land before 0.22 ships.

On the other review threads: the android.yml env thread and the apple.yml env thread still apply (first finding). The private-package thread is marked resolved but still applies (second finding). One lower-priority thread still applies: the temp directory and one non-blocking resilience note on registry.ts:290-297 (thread). These do not apply, so please resolve them: execute.ts:203 (runner.close() has only sequential callers) and the fixed-at-head threads r4572488729, r4572488730, r4572488740, r4572488744, r4572488752, r4572488753, r4572488761, r4572488765, r4572488769, r4572488773, r4572488780, r4572488788, r4572488796 (the guard works; the union above would remove it) and r4572513397.

The pull_request CI, Linux, Coverage and Integration lanes pass. The pending action-required check likely comes from the open resilience note on registry.ts, a file this PR changes. The Android and Apple push runs fail only because of the invalid workflow files above.

Before merge, the smoke workflows must start and pass, @agent-device/maestro must be installable from the release, and the live packed-plugin Android and iOS replay output must be attached.

For scope: I reviewed the delta 1a9eadd..0158fac and did not run tests, builds or the packed smoke, so the 46 focused tests and 6 packed cases are unverified by me. The workflow rejection is inferred from GitHub's documented context rules and the run records, not from an annotation text. I did not check whether 0.22 is cut from this branch before the package can be published, and I did not re-read the resolved threads on files outside the delta (replay-plugin-setup.ts, vitest-tmpdir-global-setup.ts, host.ts).

@thymikee

Copy link
Copy Markdown
Member Author

New exact-head local validation for the unpublished Maestro prerequisite pair, run once sequentially with direct pnpm 11.17.0 frozen setup/build at each candidate:

Head / invocation Raw exit Passed Skipped Exact/unchanged
ebedf50e755f1657b5cb6add0bd1d50d174982ed affected 0 42 25 true/true
ba0d2ea1aa6eaff0f4ee97bf7b7ef90e905aa9b7 affected 0 42 25 true/true
ba0d2ea1aa6eaff0f4ee97bf7b7ef90e905aa9b7 full 0 41 25 true/true

Both affected runs retained FAIL-OPEN full selection and default origin/main d78be6c, merge base d38e35e. Actual 0.22 base0081 remains separate. Counts overlap; skips are not hosted/manual/live passes.

Parent verified all 286 manifest hashes in the evidence directory and archive, all 154 original/copy receipt/log files byte-for-byte, clean initial/final exact heads/fingerprints, pinned setup and all three raw exits. Receipt SHA256 values, in table order: ef604843409972f42a37c62315adc258fce9fda9b356e3c3da3034f3181f5af1; 2ef71733a41556dee838eabaca08ccce868fd7a75572360e79a72b541a08cd2e; 28eb4bf1482a84b629e107ed574bf85fd72b29d819689e64b7029b3f3efe0515. Archive SHA256 930afe5e69482c0bf9b6ce4894def06e321d52f8fd58d981098a7eaa6797d5c7.

The archive is verified but UNSEALED: optional /bin/chflags was unavailable, so the controller exited 1 after verification. No sealing retry occurred. This does not change the three gate exits. No printed lock wait; actual wait unmeasured.

Independent source review remains running. Both candidates are unpublished; published PR remains0158. Historical DAD/C27/metadata/timeout failures retain their original status. No hosted, Linux-causality, live, performance, private-package/npm, ADR or merge clearance is inferred.

@thymikee

Copy link
Copy Markdown
Member Author

Exact local admissions for the NEW atomic pair, each run once sequentially:

  • PREP feb7c25 affected: raw0, 42 passed/25 skipped, receipt SHA256 39337f27e6409ec00bd5a4b6652dbab2bc79ba9eb4c3877c44af4f073e0b7fcf.
  • CONSUMER c659cfbd651ee023e11a8b06f37b6861601e59fc affected: raw0, 42 passed/25 skipped, receipt SHA256 26538be2de62f46e7e01249ba9b5b0ddea2c5fe5baf81fc28c758a5cd53b3522.
  • Same CONSUMER full: raw0, 41 passed/25 skipped, receipt SHA256 aa0baeeb64712b7099cddd882d5e9945181b46a81bcf9d3a172d1824ffdde47e.

All receipts exactCommit=true, checkoutChanged=false with clean matching initial/final fingerprints. Both candidates passed direct pnpm11.17.0 frozen install/build. Both affected runs retained full FAIL-OPEN selection and default origin/main d78be6c, merge base d38e35e; actual0.22root0081 recorded separately. Restored plugin rank4 remained unchanged and layering passed. Counts overlap and skips are not hosted/manual/live passes.

Parent independently verified all232 manifest hashes in directory and archive and all156 original/copy file pairs, exact refs/results/clean fingerprints, pinned setup commands and Git rank4. Archive SHA256 8e42cd6dbf3da43430323bf80f738764e1762457f905cbaf02c9488f98927e8e. Evidence is VERIFIED, UNSEALED; no sealing attempt. Controller exit0; separate verifier initially had a syntax error before execution, followed by direct verification exit0. No gate rerun, source repair or gate relaxation occurred. No zero lock-wait claim.

New independent source recheck is still running. Both layers are unpublished; public PR head remains0158. Earlier endpoint passes and intermediate sequencing failures retain their original SHAs; historical DAD/C27/Fallow/R11/TSref/frozenlock/E8timeout failures get no causal waiver. Hosted/live/current-client/ADR/performance and private-package publication obligations remain open; no merge clearance.

@thymikee

thymikee commented Nov 14, 2026 •

Copy link
Copy Markdown
Member Author

Root9f replay-stack local admission — 2026-11-15T13:18:36.516082+00:00

Current unpublished family: f3e9ddbd → f85c7ddb → 1d044457 → 5e6257c7 → 5a335ab510bbadac508b60240a44accb2decdcb0, on 0.22@9f6435f0 / main 7848a1a9. Published PR heads remain c3b/924/6d0/61ae/ce66; nothing was pushed or merged.

Independent source review is clear, including the repaired TOP delta. New TOP pinned Node26.8.1/pnpm11.17 frozen install, build, full format, affected and full each ran once/raw0. Affected passed42/skipped25; full passed41/skipped25; both exactCommit=true/checkoutChanged=false with equal clean fingerprint b9d590bf58b94f6df5e6ed75477439d06e8d63f08a1c86bfef24d4b97f975087. All77 commands used one captured ENV with AGENT_DEVICE_VITEST_MAX_WORKERS="2"; the wrong key was absent. All9428 baseline refs, tracked source and logical index stayed unchanged. Counts overlap.

Named exact-head receipt SHA256 values (individual passes, not an uninterrupted protocol claim):

  • PREP affected f3e9ddbd: 6268ab5fca04b99fa1ee60b6ff7abd5b3551583d2dd6942ac86c65facff7dc74.
  • C3 affected f85c7ddb: 2fcf9591dc218cbde00a60a3708673871c0dc83def75229605020af17910a9d5.
  • C3 full f85c7ddb: 482bbb2cbd020eb05b37d82de3925d8605001d227c36ebef72fec0e7367851f0.
  • DATA affected 1d044457: 7ee13e59056f9686e9b7a554e2c7b4f908be074df36be55179cad4628e3d3e81.
  • REG affected 5e6257c7: 28da1eb7efbac3f7f115ce41db04b45ad2de9925bf6591893ff01c75018aa211.
  • Repaired TOP affected 5a335ab5: 32d9ab9ea4dffad214a4041acc92a61dd54c5715ad537c7eb63a50556a5f1892.
  • Repaired TOP full 5a335ab5: 5b0c9667755d4e78d0121348b4be1c3a173cf0f0771f0b3cadfd2b12eb051996.

TOP physical archive SHA256 208c2bc8bc2f52d68cc4d9219fbeda86fb856b647c19d1389d4a8137b662b17e (79,996,063 B), VERIFIED UNSEALED: all archive bytes/inventories/modes, 6,112 source/blob pairs and complete 52+50 original/copy receipt files checked independently. All6,257 recorded metadata/copy pairs match; 6,252 current original pairs match. Five historical raw-index captures retain their own recorded hashes while the current index matches the final capture; byte-change cause is unknown. All seven complete logical-index logs match the committed tree. Initial/final snapshots cannot exclude transient edits; bootstrap UTC and lock wait are unmeasured/unknown as recorded.

The earlier TOP730f affected FAILURE/full7 UNRUN, wrong-worker-key pre-setup stop, REG output-collector stop, DATA parent-ref stop and predecessor varying whole-ENV digests remain retained. New TOP passes do not waive those historical protocol qualifications. No unchanged-head broad check was repeated to erase them.

Publication remains blocked on acceptable owner disposition of Node22.13 developer packaging and predecessor ENV-protocol limitations. Both optional packages stay private until maintainer non-latest placeholders, release.yml job npm-publish trusted publishers and usable publication are verified before any PREP/plugin merge. Native ADR3871 acceptance, canonical fixtures/NavARM/nativeCatalog/HTTP recording/current-client/platform-remote/live/performance and nightly CodeQL4HIGH remain separate open obligations. No new-head hosted or release clearance is claimed.

Earlier records follow unchanged:

Daily integration maintenance — 2026-11-15T08:40:46.252921+00:00

Only 0.22 was rebased and pushed: 0081f616a2cfd1088511c4e172db262a08e9529a → 9f6435f039cac6458b24e755e2486f457a627856, onto exact main@7848a1a9283fe36626f5cb6bb62423b9a084f09d. Eight integration commits remain: seven range-diff equals and one host-test inherited-context-only difference; all added/removed line sequences match. The explicit old-0081 lease was accepted. No tags or release were created.

At new root9f, affected passed42/skipped25 (receipt SHA905314e9fa2ac7828d91e2c38cd12eb7d8c10203e1a9eb86f5a1ecf7686a79f4); full passed41/skipped25 (SHA3eeaff8d7d1dfb61177cfe211cdbaba98d9dd019f08c1ff78d091bcbff88fce0). Counts overlap. Both exactCommit=true/checkoutChanged=false, matching clean fingerprint dbe87e7f1276093a16073183ec25992340d71171a5f5906b94658e2f76780fdb. Node26.8.1/pnpm11.17 frozen setup and full format passed; format changed no tracked bytes. Every skipped owner remains a separate obligation.

The initial root affected attempt FAILED37/25/1/4 at agent-guidance ENOENT for a manifestless ignored packages/replay directory; full was not run then. The complete owned generated directory was moved to preserved storage with byte/mode/symlink inventories equal, without source/gate changes. Focused guidance then passed5/0; its environment differs, so no identical-env pair is claimed. The failed receipt remains retained. Subsequent root admissions above passed. Validation-only archive e02792b098adf5fc2c44137919ebbe34ff670a1b2c63db28baab9374cdba7f68 has211 verified regular files including its root manifest, UNSEALED.

Published PR4555/4505/4534/4535/4537 heads remain c3b/924/6d0/61ae/ce66; A5 PR4539 remains8c. The five local corrections 8edf/3c467/03b488/27c7/3ed5 remain unpublished and based on OLD0081: their source/Node26 proofs do not admit a new-root restack. Node22.13 developer packaging owner disposition remains required. PR4555 now displays196 files/4907 gross and A5 displays207/5361 through old ancestry; their reviewed original layers were38/794 and48/1246. Upper immediate PR layers remain unchanged. Isolated A5 refresh is next; plugin refresh remains blocked, with new-head review/validation still required. Accepted unpublished findings stay open. Private publication/native ADR/canonical/current-client/live/performance/security blocks remain. Root local success is not hosted/live/npm/merge clearance.

All previous evidence below retains its named SHA and original scope.


Current five-head Node26 local validation verified; publication still blocked — 2026-11-15

The completed sole validator ran all seven canonical invocations exactly once on verified Node26.8.1/pnpm11.17.0. Every receipt is raw0/passed/exactCommit=true/checkoutChanged=false. Five affected runs selected all67 under FAIL-OPEN; two full runs selected66 excluding related. Every25 skipped owner/reason is retained; counts overlap and are not summed. Default origin/main semantics remain separate from actual0.22 layer diffs.

  • PREP affected 8edf266044f052e9252372110395f8a00a7ca3f9: 42 passed/25 skipped; receipt 11f07d87ebe8e6c4064b0ea33ffb1f3be610f7d703ef2e12cf636352d1ad9a2a.
  • C3 affected 3c467c4ee69d38459fffbac7c4b810b732a42f70: 42 passed/25 skipped; receipt 719e9372521937ddab67a0883c37344c0e36e369a7f218333ed19e811cf577c5.
  • C3 full 3c467c4ee69d38459fffbac7c4b810b732a42f70: 41 passed/25 skipped; receipt f68726f80cad9e71737b41a073bc46f5de14a6e8657e8a2f20ab459e61b8541c.
  • DATA affected 03b4881424111a54c896d91b1bf68c49d977f911: 42 passed/25 skipped; receipt 4672f9eca5907105971c3242f3e33fc27b26b6b2da4a7be8ebd2b9169a413850.
  • REG affected 27c7cea6487abfc4b7fcb67865da1fd30dc431eb: 42 passed/25 skipped; receipt 1a2da4b312baec3098fc331ef8baf715deaf6c77f8d39cd88decee8682f1e5a6.
  • TOP affected 3ed5ed7891c5ff66b66a0eca27a26a5f9a030d3f: 42 passed/25 skipped; receipt 51779517a48b7c201c0f9f31c67d7efcaec37f8e32e5c89b3ec4f5507d2b1dd7.
  • TOP full 3ed5ed7891c5ff66b66a0eca27a26a5f9a030d3f: 41 passed/25 skipped; receipt 680d481e21f41d2ab72c466a0f6a2bb28f822d927e2beaaf07ae99ebe8713aed.

PREP's one additional check:package -- --pack-destination … invocation exited0 and retained genuine agent-device@0.22.0-dev,1,684,397 bytes/704 tar entries, SHA256 e27c08f3e35d4665f7434cd55bc6cecb856c6712b1ac852b597ca6f82e31202a. It ran the actual artifact case and core probe, including all15 exports from a clean install; this is supported-toolchain retention proof, not release:prepare, publication, floor or live proof.

C3 client-load pair: exact single candidate-filter removal produced intended RED1 (marker true versus false after typed missing-plugin assertions,1failed/3passed); exact source restoration yielded GREEN0/4passed. Argv/cwd/environment/test bytes/logical index match. Committed test SHA256 e5d3370d6c36abf6b6652d9805474b6545386420602978c4be79ed460abab94a; restored production a959647933e81a5204019f3a347a672e0364f3eac19675fb663fce56a5285d57. This proves client module-load selection, not provider startup or live behavior.

Parent independently verified complete archive inventory/bytes/digests:3,378 manifest entries comprise3,311 files+67directories; rootmanifest is an additional file (3,312 regular archive files total). All1,687 internal original/copy entries plus360 external-original entries and seven complete receipt trees match, including nested manifests. Archive UNSEALED SHA256 6c18d71305f1c9ed0406876e310b8654755f4391d82223c51b677a8a165f15f2. Final checkout clean at3ed5; four checkpoint additions recorded separately, no protected refs moved/deleted. Bootstrap discovery UTC was not captured. Evidence-only verifier interruption/raw130 and subsequent successful unchanged-archive streaming verification are retained; no setup/test/gate/mutation/semantic command was repeated.

Source admission was already independently cleared at these same five frozen heads. Combined SOURCE + SUPPORTED-TOOLCHAIN LOCAL evidence is now complete; Node22.13 developer packaging remains OPEN/BLOCKED, with original hidden stderr still unknown. No floor waiver, tool/dependency/floor/lifecycle change or alternative floor protocol is inferred. Published c3b/924/6d0/61ae/ce66 remain unchanged; new candidates remain unpublished. No CI rerun, device action, public-thread resolution, new worker, push or merge. Private-placeholder/trusted-publisher/usable-publication, native ADR, canonical/current-client/live, performance and security blockers all remain separate.


Earlier evidence (named heads and times only)

Guard-family independent source recheck clean; runtime blockers remain — 2026-11-15

Completed Apex MEDIUM recheck at TOP 3ed5ed7891c5ff66b66a0eca27a26a5f9a030d3f reports NO FINDINGS requiring action in the actual guard repair and its interactions. Report SHA256 214bf5e231f24dc15e9f22a93b23e9e8e8ee34f085877c60092e89a7dc4f2606. Parent read the complete final result/report and independently verified hardcoded guard equality at all five new heads, guard-only1+/1− endpoint diffs, unchanged package/lock/CI/catalog bytes, exact DATA/REG patch preservation, correct immediate scopes and the clean final source-review checkout. F-A is fixed locally and independently reviewed. F-C retain-observation disposition is supported by actual build-first callers; no extra guard or speculative repair is needed. No source changes followed the review.

This completes current source admission using the prior0839 source review plus the verified guard-only delta and this new recheck; it does not borrow an old runtime pass. Current frozen unpublished heads remain 8edf266044f052e9252372110395f8a00a7ca3f9 / 3c467c4ee69d38459fffbac7c4b810b732a42f70 / 03b4881424111a54c896d91b1bf68c49d977f911 / 27c7cea6487abfc4b7fcb67865da1fd30dc431eb / 3ed5ed7891c5ff66b66a0eca27a26a5f9a030d3f. Scope remains five files/45gross versus publishedce66; no new package/dependency/export/gate/baseline/floor change, both plugins private.

The SAME sole supported-Node26 validator continues its seven exact invocations, retained-core-tarball and client-load proof contract without running reads, steering or duplication. Node22.13 developer packaging remains BLOCKED; the direct nested diagnosis failed in tsdown’s unrun loader, while original node:test-hidden stderr remains unproven. Source review does not clear that failure, publication, hosted/live/current-client, ADR, performance, security or merge readiness. No new worker, source push, CI retry, public thread resolution, device action or release. Existing published heads and watches remain unchanged.


Earlier evidence (named heads and times only)

Nested-pack diagnosis verified; supported-toolchain lanes remain separate — 2026-11-15

The completed one-shot diagnosis at PREP15036137 captured raw exit1 in Maestro prepack: tsdown0.22.14 failed to import its optional unrun config loader. Its native loader requires native TypeScript support plus module.registerHooks; installed dependency source documents registerHooks from22.15, while this capture used verified22.13. The optional unrun peer is absent. Root tsdown/lock inputs predate PREP (SOURCE inheritance only), while the Maestro prepack lifecycle is preparation work. This is observed failure of the new direct nested command, not proof of the original node:test-hidden stderr or unchanged-base execution. New environment/cwd/scratch differ; original hidden stderr remains UNPROVEN.

Parent independently verified536 archive files INCLUDING rootmanifest,535 manifest entries, and316 original/copy pairs, including nested old manifests. The child's535-file wording counts non-root entries, not the total archive. Archive UNSEALED SHA256 dece4fb159ff063ff5c9b0ce224df8af6eafdd1979726df2dc9bc9940e2648bb. Setup0/0/0, exactly one pack1, no tarball/retry/repair; clean PREP/index/source/existing refs preserved. No whole failed-sequence retry, dependency upgrade, floor raise, optional-peer install or lifecycle skipping.

Node22.13 developer packaging remains BLOCKED. A NEW sole Codex validator owns the already-required supported Node26/pnpm11.17 lanes on the guard-fixed local family8edf/3c467/03b488/27c7/3ed5: fresh setup at each head, seven ascending exact canonical invocations, one explicit retained-core-tarball destination check and one client-load RED/GREEN. No Node22 retry or substitute floor claim is authorized. Any Node26 passes remain independent and cannot waive the original floor failure. The Apex MEDIUM guard-family source recheck continues unchanged; original A1 remains queued and unpolled. Current new-head results are still pending.

All published heads remain c3b/924/6d0/61ae/ce66. No source push, CI retry, device work, release or merge. Required Node-floor/private-publication/native-ADR/canonical/current-client/live/performance/security blockers remain open; floor policy/toolchain changes need a separate owner disposition.


Earlier evidence (named heads and times only)

Independent source finding accepted; guard-only repair restacked — 2026-11-15

The completed cold Apex MEDIUM final-family review at TOP0839dcffa45946909c3a667d80937713259cfa1f is NOT CLEAN: one major blocking finding F-A. The package-command reorder/child flag is correct, but the unchanged exact command-array guard in test/repo-guards/npm-package-scripts.test.ts contradicted it. Parent independently verified the mismatch at all five old candidates (and equality at all five published heads). Report SHA256 bda2937262305e8b185086cca9d6163847b602822abaf9f6726bbea652cd9493. The review separately confirmed client selected-loader minimality/coverage, child flag necessity, and native-vs-format docs/F2 dispositions; that is not whole-family approval.

The minimum parent fix synchronizes the guard's two literal expected segments without relaxing its assertion or removing artifact/smoke cases. Actual focused owning-suite RED: 1 failed/8 passed at publishing cannot skip the package gate; repaired GREEN: 9 passed. Package bytes/argv/cwd match, but the two environment digests differ and test bytes change only the expected-array repair: parent feedback, not an identical-environment mutation or canonical admission. Fresh pinned PREP install/build and full-repository PREP/TOP format passed. No new broad gate ran.

New frozen LOCAL UNPUBLISHED family: PREP 8edf266044f052e9252372110395f8a00a7ca3f9, C3 3c467c4ee69d38459fffbac7c4b810b732a42f70, DATA 03b4881424111a54c896d91b1bf68c49d977f911, REG 27c7cea6487abfc4b7fcb67865da1fd30dc431eb, TOP 3ed5ed7891c5ff66b66a0eca27a26a5f9a030d3f. Every corresponding old/new endpoint differs ONLY in the guard (1+/1−). C3/TOP guard conflicts were reconstructed from exact original files plus this bounded correction; every layer's test-path suffix is preserved. DATA and REG patches are byte-exact. Immediate layer scopes remain 38/794, 104/2472, 60/762 (no-renames64/4892), 12/385, 74/1882; default-main TOP remains 358/9958. Publishedce66→newTOP is now five files/28+/17−=45gross. Old aliases are preserved.

A NEW Apex MEDIUM source recheck owns this repaired family. The separate Node22.13 nested-pack diagnosis keeps its original PREP1503 assignment; no running reads/steering. Prior package restored GREEN remains FAILED with cause unproven/no retained core tarball; all seven canonical gates and client-load pair remain NOT RUN, with no new-head transfers. F-C minor bare-unbuilt check loses the early build diagnostic after the swap; no checked-in lane triggers it because setup/build precedes package. The observation is retained for independent recheck, with no speculative extra guard, lifecycle skip, floor raise or dependency upgrade.

Published c3b/924/6d0/61ae/ce66 are unchanged. No source push, CI retry, public review-thread closure or merge. Required package/source/local/hosted/live/private-publication/native-ADR/current-client/performance/security blockers remain separate and open.


Earlier evidence (named heads and times only)

Current-family validation stopped at PREP package GREEN — 2026-11-15

The completed sole validator stopped exactly as required at PREP 15036137c69b9601d933bb50fba24e089aa53e35. Actual Node22.13 RED exited1 at the intended host-kit command.ts ERR_UNKNOWN_FILE_EXTENSION after removing only the child strip-types flag. Restored identical-env/argv GREEN also exited1, now inside the real artifact case at nested pnpm pack. Its node:test report omits captured nested stdout/stderr; underlying cause remains UNPROVEN. The core probe never ran, retained destination is empty, and there is NO retained-tarball or successful package-floor proof. All seven canonical invocations and the C3 client-load pair are NOT RUN; zero exact-commit receipt admissions. PREP install/rootbuild/explicitMaestrobuild passed under Node26 and do not prove the Node floor.

Parent independently verified all423 archived file bytes, 422 manifest entries and 210 original/copy byte pairs. Archive UNSEALED SHA256 dbf3994cbb6b190bedd2494fc15852d6d2b28e1c1638425747c858823c5dbf18. package.json exact saved bytes/index and clean PREP fingerprint were restored. Both pair legs share argv/cwd/scratch/test bytes/environment digest; this is expected RED plus FAILED restored GREEN, not a successful RED/GREEN proof. No retries, source repair or later gates were performed. Discovery helper's corrected archive-path read occurred before setup and is separate from the actual failure.

A new bounded diagnosis task will expose exactly the failing nested pnpm pack stdout/stderr in its own prepared PREP checkout after fresh pinned setup, with no source changes or full check:package retry. Installed tsdown0.22.14 declares Node ^22.18.0 || >=24.11.0; that is a hypothesis to distinguish from package/lifecycle or pnpm failure, not proof of the original cause. No dependency/floor/lifecycle workaround is authorized. The independent Apex MEDIUM source review at TOP0839 continues unchanged; original A1 remains queued and unpolled. Current final family remains 1503/9025/3ac/2390/0839, unpublished and unadmitted. Older f648/23be passes stay at their named heads. No source push, CI retry, device action, publication or merge; all package/live/ADR/current-client/performance/security blockers remain.


Earlier evidence (named heads and times only)

Older docs validation verified; current-family validator running — 2026-11-15

The completed Codex docs-only validator ran REG f64803d5acf173343ccd5be6977d6f5a6d11eda4 affected and TOP 23beaeeba83228ada4d108487cd446b7345c4ba2 affected/full exactly once: 42/25, 42/25 and 41/25, respectively. Every receipt has exactCommit=true and checkoutChanged=false. Parent independently verified all 352 archive files, 351 manifest entries and 177 original/copy byte pairs; archive is UNSEALED. SHA256 82c1fec2bce6716cf625beb4bf5cbf1424dbd2777524504a458d06e3584d5774. Raw receipt SHAs: 3ca1f143b5b804f5ed81a25de2cb1e564cddd0cdb5d76e97f596026e0389d7ff, 2c8791c7da7218978ead442a56f6696445080d20422a1bdc28afb69947e32e0e, 6a1467b7685f9317ce3566194953a7ae1acaee774156bbdceb59accbc8e2d4e1. Counts overlap; skipped owners remain separate. Clean final23be/tracked bytes/logical index unchanged; 20 ref additions and no protected moves/deletions. Read-only missing-path/glob incidents occurred before execution; all nine setup/canonical commands exited0. Lock wait is unmeasured. These are OLD named-head results, not current-family proof or authority to push the older pair.

Current frozen LOCAL UNPUBLISHED family remains PREP 15036137c69b9601d933bb50fba24e089aa53e35, C3 9025146e239005dac2b838f9edf7e21d40d92ff5, DATA 3ac285a014d722644523e57f69cee26081e12b45, REG 23908e310de76ba9e36b699f6db13a30360283ca, TOP 0839dcffa45946909c3a667d80937713259cfa1f. A new sole Codex validator now owns actual Node22.13 package-child RED/GREEN with retained core pack destination, the unrelated-client-module load RED/GREEN, fresh setup at each head and seven ascending exact canonical invocations. The independent Apex MEDIUM final-family source/prose review continues unchanged. No running reads, steering, fourth worker, result transfer, source push, manual CI rerun or device action. Existing published heads and accepted-finding threads stay unchanged. Required source/local/hosted/live/private-package/ADR/performance/security lanes remain separate; no merge clearance.


Earlier evidence (named heads and times only)

Docs review completed; final-family recheck running — 2026-11-15

The completed independent Apex MEDIUM docs-only review at REG f64803d5acf173343ccd5be6977d6f5a6d11eda4 / TOP 23beaeeba83228ada4d108487cd446b7345c4ba2 approved the native-versus-format setup correction with no blocking findings. Report SHA256 c4b50756ed5925fd64cdca063addc245d844a97cfb3c65b9f686cc686dc8acdd. This is source/prose evidence at those heads, not a canonical, hosted, package, ADR or merge clearance.

One optional terminology refinement was accepted: “the object returned by create()” replaces the ambiguous “runtime” in the native paragraph. The suggested reserved-ad rule was declined as an unsupported new contract; optional completeness text was declined; truthful private/unavailable disclosure stays until actual maintainer publication. ADR acceptance remains open.

Current LOCAL UNPUBLISHED family is PREP 15036137c69b9601d933bb50fba24e089aa53e35, C3 9025146e239005dac2b838f9edf7e21d40d92ff5, DATA 3ac285a014d722644523e57f69cee26081e12b45, REG 23908e310de76ba9e36b699f6db13a30360283ca, TOP 0839dcffa45946909c3a667d80937713259cfa1f. The new REG/TOP change from 272f/66d0 is one public sentence; production, package-command and test bytes are unchanged. The old TOP layer is preserved by range-diff '='. Published ce66 → current TOP is four files/27+/16−. Existing family/layer/default-main scope disclosures below remain unchanged.

A NEW independent Apex MEDIUM final-family source/prose recheck now owns the package-argument/Node-floor, selected-client-loader, documentation and wording interactions. The existing serialized Codex validator remains frozen at f648/23be; no running reads or steering. Its eventual results retain their named-head scope. Actual supported-floor/retained-pack-destination/client-load regression proof and seven new-family exact gates remain pending. No old-head result transfers; no source push, CI rerun or merge. All private-package/native/canonical/current-client/live/ADR/performance/security blockers persist.


Earlier evidence (named heads and times only)

Client replay scope finding accepted; local candidate family updated — 2026-11-15

PR #4505 Sentinel finding4576579437 is accepted for evaluating unrelated provider modules/factories in client replay scopes. Import errors are caught into inventory, so the claim that an unrelated throw directly fails replay is not established; unnecessary evaluation/side effects/I/O are the concrete defect. Store metadata discovery still reads the store. The local correction resolves the vetted backend package once and filters candidates before registry loading. Embedded resolvers and the daemon's complete registry remain unchanged.

The existing missing-plugin regression is extended with an unrelated store package whose module evaluation writes a marker and throws. It retains the typed missing-plugin/package/install-command assertions and additionally requires no evaluation marker. No test was deleted, no new abstraction/dependency/wire/SDK shape was added. The intended unfiltered RED → filtered GREEN has not run yet. Full repository format passed; no new-head install/build/focused/canonical or performance pass is claimed while the original serialized validator remains active.

Updated LOCAL UNPUBLISHED family:

  • PREP 15036137c69b9601d933bb50fba24e089aa53e35 (unchanged package-command correction)
  • C3 9025146e239005dac2b838f9edf7e21d40d92ff5
  • DATA 3ac285a014d722644523e57f69cee26081e12b45
  • REG 272f18d19e3583846a943e7de1e0d3821dc47ed5
  • TOP 66d0daa23ccac8f92167bf40b3b2d04b269dfcf8

C3 correction is two files/20+/13−. Every corresponding old/new C3-through-TOP endpoint differs only in the registry owner and existing test file; package-script and documentation corrections are byte-preserved. C3 immediate scope is 104 files/2,472 gross; DATA remains 60/762 (64/4,892 without renames); REG 12/385; TOP 74/1,882. Separate default-main TOP comparison is 358 files/9,958 gross. All older local aliases and published c3b/924/6d0/61ae/ce66 remain unchanged.

Independent final-family source review, client-load RED/GREEN, actual Node-floor and retained-pack-destination proof, and seven exact gates remain pending. The existing f648/23be documentation workers keep their frozen assignments and their future results stay named-head evidence. No new/cancelled/steered/polled worker, fourth reservation, source push, CI rerun, device action, release or merge. Accepted package-command findings and all private-package/live/native/canonical/current-client/ADR/performance/security blockers remain open. This finding stays open until reviewed, validated publication.


Earlier evidence (named heads and times only)

Package-command findings accepted; local five-layer correction — 2026-11-15

PR #4555 Sentinel is action_required at published c3b5411. Both findings are accepted. The current check:package suffix takes release:prepare's forwarded --pack-destination away from scripts/check-package.ts. The prior strip-types dismissal was incomplete: the documented development floor includes Node 22.13–22.17, which still needs the explicit child flag (default stripping arrived in Node 22.18).

A parent-owned local correction runs the existing artifact tests first, explicitly forwards --experimental-strip-types to their child, and keeps the core package probe last to receive forwarded release arguments. No new wrapper/gate/test deletion or baseline change. Every layer's artifact/smoke test suffix is preserved. New LOCAL UNPUBLISHED family:

  • PREP 15036137c69b9601d933bb50fba24e089aa53e35
  • C3 282db0e5c9948fb4dfb9613321c288bc1e795e06
  • DATA 82428eb3945c7cd89b6226be6a3e7b390d760f54
  • REG a68e992ce52cade6a180627e4b9b98ea62aa9b9e
  • TOP 3c7f1cc9fa9ce359df63da6ba5d1a34a3e55287d

Each endpoint differs from its corresponding preceding local head only in the check:package script value; the new REG/TOP documentation correction is byte-preserved. Immediate family sizes remain PREP 38 files/794 gross; C3 104/2,467; DATA 60/762 (64/4,892 without renames); REG 12/385; TOP 74/1,882. Default-main TOP comparison is 358 files/9,953 gross.

A standalone pnpm 11.17.0 argv-recording probe used actual PREP and TOP script strings: old argument ownership FAIL, corrected ownership PASS; all raw pnpm exits 0 because this fixture records argv only. This is not an actual package/release/floor/canonical test. Full repository formatting passed; no new-head install/build/test/gate claim exists. Independent review, actual supported-floor/package-destination proof and seven exact gates remain pending. Existing documentation source/validation workers remain frozen at their original f648/23be snapshots, without polling or steering; their eventual results stay named-head evidence.

Published heads remain c3b/924/6d0/61ae/ce66. Source admission of the family is blocked pending the correction. DATA 6d0 hosted 18 success/2 skip and REG 61ae hosted 15 success/2 skip/1 neutral remain valid named-head results, without transfer. No bypass of Sentinel action_required, no CI rerun, publishing, release, merge or device action. Both finding threads remain open until reviewed, validated publication. All private-package/live/native/canonical/current-client/ADR/performance/security blockers persist.


Earlier evidence (named heads and times only)

Current registry documentation correction — 2026-11-15

Sentinel's plugin setup documentation finding in PR #4535 is accepted. The published page describes only format-backend replayBackend.create(host); the native ad validator requires execution.create() and rejects mixed shapes. A parent-owned documentation-only correction is committed locally at REG f64803d5acf173343ccd5be6977d6f5a6d11eda4 and carried into TOP 23beaeeba83228ada4d108487cd446b7345c4ba2. Both endpoint deltas are one documentation file, 6+/2−; lower PREP/C3/DATA heads are unchanged. REG is 12 files/385 gross; TOP remains 74 files/1,882 gross. Old TOP production-layer patch is unchanged by range-diff. Independent Apex MEDIUM prose/source recheck and three exact-head gates (REG affected; TOP affected/full) are running. No new candidate pass or source approval is claimed; neither candidate is published. Published REG/TOP source admission remains incomplete pending this correction.

At the unchanged published REG 61ae277, current hosted workflows are nonfailing: 15 success / 2 skipped / 1 neutral. The iOS Settings and Android clickable-first smoke tests passed; changed-line coverage enforced 38/39 = 97.44%, threshold 70%. CI merge tree equals that head. Full mutation sweep was skipped; selected decision-kernel mutants passed. Earlier cancelled workflows and historical f390 failures retain their original state and causal qualifications. The native-create retry finding is already fixed/published/resolved; its notification does not request another fix.

Source/local/hosted/live/package/security/ADR/performance lanes remain separate. Both plugins stay private; maintainer placeholders, trusted publishers and usable publication plus outstanding live/canonical/current-client/native/ADR/performance/security requirements remain merge blockers. No source push, manual CI rerun, device action or merge from this update.


Earlier evidence (named heads and times only)

Published final family — source clear and seven exact local passes

The five-layer repair round is now published atomically, using explicit old-head leases: #4555 c3b5411 → #4505 924a0bd → #4534 6d011ba → #4535 61ae277 → #4537 ce66416. T3 linked all five PRs in the next tool call. API-only gh stack link confirmed the existing five-PR stack and correct base chain. All remain open and ready for review; auto-merge is off. Root 0.22 remains 0081.

Independent Apex MEDIUM source review found no findings in the actual final family, including host-owned iOS origin refusal and native-create recovery. Source report SHA256: dad96bcb54c84a71998e8a1ae0af8a4bd239aed8e1a82071bcc83a39fd1e187b. Runtime effort is unobservable; this is a source opinion, not CI/device proof.

The separate validator completed exactly one committed origin regression pair and all seven canonical invocations, without retries. Each receipt exited 0, exactCommit=true, checkoutChanged=false. Affected selected all 67 checks through FAIL-OPEN; full selected 66 and excluded related tests. Every lane retains its 25 skipped owners. Counts overlap and must not be added as unique coverage.

Invocation Exact head Result Receipt SHA256
01-PREP-affected c3b5411763de2639799d991323f6059005803e25 42 passed / 25 skipped c72532bec73c34432bc0509f25f232b4993df752616aa111077fd4eaf2dfc187
02-CONSUMER-affected 924a0bd3570cb5ff45751d4b3b1a522415794064 42 passed / 25 skipped 7a27a2aab46efa7df5b609447e3e2b48094bafc25bd50066f36fdc7f58a62fa3
03-CONSUMER-full 924a0bd3570cb5ff45751d4b3b1a522415794064 41 passed / 25 skipped 279b9c909a75abee6b94162694dbcf889af5629d441c10c4cb18bbb2bc5f6099
04-DATA-affected 6d011ba80db214fa1adeb5373a69bbbde262b9c6 42 passed / 25 skipped 2231b3e9dc97beb9287d4bbb778a40a5a402051ab6a4382dcbbe0957873bc987
05-REG-affected 61ae2772d4ea72da9544abfdd305705e3edd51fd 42 passed / 25 skipped ca0b49cd9cdcf46ee526a8d5f0ce11966f99b5f8916e8809698cb510273a424c
06-TOP-affected ce66416f9c516b3839d3bf8620eb0f77a612b725 42 passed / 25 skipped f3355a358b09c5dfba8786ec33d6f83bcdd419957b957631968b896704b94359
07-TOP-full ce66416f9c516b3839d3bf8620eb0f77a612b725 41 passed / 25 skipped 06ee33eabcd0ffb980e6ff2fd64c1d67e2cbc4412720c49e51c5957dbc370521

The bounded host-assertion mutation produced RED exit 1 at committed helper line 98, “derived tap must refuse before input,” true !== false. Exact host-byte restoration followed by identical argv/captured environment produced GREEN exit 0 for all six derived/missing/reported tap/scroll combinations. The 18 nonempty built-plugin files stayed unchanged during the pair. Helper SHA256 d03fe1df076ab182b179f44d82beca1720bf42ebc108a5fafd938153b4138959. This uses production-source host acquisition/presentation/snapshot assembly with a separate built plugin and intercepted native input; the canonical package cases additionally pack/npm-install it. It is not full-packed-core or live-device evidence, and does not identify historical CI dispatch coordinates.

All five fresh frozen installs/builds used direct pinned pnpm 11.17.0, Node 26.8.1, Python shell=False and workers 2. T3 request 2026-11-14 23:46:18.179 UTC → start 23:46:19.479 → completion 2026-11-15 01:11:12.071; elapsed 1h24m52.592s and request-to-start 1.300s. Lock wait remains unmeasured. Configured inherited Codex/gpt-6.1-sol/high/default; executedModel null, no independent runtime-model attestation. Final checkout clean at ce66; logical index restored; four shared checkpoint additions and no existing protected-ref moves/deletions.

Parent independently verified every one of 579 archived files/578 manifest entries, all 360 receipt original/copy byte pairs and seven complete receipt inventories, 10 committed-source copy pairs, and 36 nonempty before/after plugin snapshots against their captured pair baseline. The latter is not a claim about later dist rebuild bytes. UNSEALED archive SHA256 aa0360512b7ec1094285fc4a762d652296943e3b886ecbababe331562838f111. Raw report: /tmp/v022-c3-c4-exact-vac9p3kv/final-report.json; archive: /tmp/v022-c3-c4-exact-vac9p3kv.tar.gz. Durable parent verification: c3-screen-origin-host-port/test-ce66-parent-verification.json.

Retained bookkeeping limits: one worker read-only lookup of nonexistent types.ts exited 1, unrelated to setup/tests. Parent verifier first expected resolvedBase on full receipts, then compared string workers “2” with integer 2; only that verifier was corrected. First versions and failures are retained; no source, receipt, archive or validation was changed or repeated. Earlier parent harness/env/stale-artifact experiments and all old-head CI/gate failures remain in historical sections below.

Fresh hosted checks on these rewritten heads are pending via the existing watches; no old-head hosted/Size result is transferred and no unchanged CI rerun occurred. Native-load finding 4575620206 is fixed in the published registry; source and local evidence do not grant hosted/live/package/security/performance/merge clearance. Both optional plugins stay private; maintainer non-latest npm placeholders, release-workflow trusted publishers and usable publication remain prerequisites before merge. Canonical/native/platform/remote/current-client/live/ADR/performance and nightly CodeQL blockers remain. No merge, npm publication, release, tag, device action or stopped-route retry occurred.


Earlier evidence (historical SHA/time scope)

Independent final-family source recheck complete — no findings

Apex, configured MEDIUM, completed a fresh strict read-only review of c3b5411 / 924a0bd / 6d011ba / 61ae277 / ce66416. It independently confirms the host-owned iOS origin refusal, native execution-create next-caller recovery, required contracts, typed repair errors, package-test boundary, scope and preservation, with no source/design/scope findings. The earlier CE51 iOS P2 is addressed in this local family. No source approval was borrowed from old heads.

Review metadata: requested 2026-11-14 23:45:33.580 UTC; started 23:45:35.199; completed 2026-11-15 00:28:10.369. Elapsed 42m35.170s; request-to-start 1.619s. Executed model/effort were not independently observable (executedModel null). Initial/final checkout remained clean at ce66416. Report SHA256: dad96bcb54c84a71998e8a1ae0af8a4bd239aed8e1a82071bcc83a39fd1e187b. Parent verified the completed final response, report, parent chain and unchanged registry/test blobs.

This is SOURCE evidence only. The separate committed-test captured-environment regression and seven exact-head gates remain pending; their running task was not read or polled. No new-head runtime or hosted pass is claimed. The five local heads remain frozen and unpublished; published b057/DDA/d373/f390/2e40 and their actual hosted failures retain their SHA scope. Native-create finding remains open until publication is verified. Private-package, native ADR, canonical/live/current-client/performance and CodeQL blockers remain unchanged.


Earlier evidence (historical SHA/time scope)

Current repair round: host iOS origin refusal (local, unpublished)

The completed independent CE51 source review accepts the clickability bridge and identifies a distinct P2: the extracted Maestro bundle reads its own private viewport map, so iOS capture-derived tap/scroll refusal can become inert. The required host assertion now preserves the existing owner and typed window-origin-unknown/dispatched=no refusal. This does not diagnose the historical Android tap or DATA native Settings failure.

New frozen local family: PREP c3b5411 → consumer 924a0bd → DATA 6d011ba → REG 61ae277 → TOP ce66416. The native-create next-caller recovery repair is retained. Published b057/DDA/d373/f390/2e40 and all hosted failures remain unchanged.

Parent feedback: pinned setup/build, full format, typecheck, lint, layering, Fallow, 17 focused tests and one actual packed/npm-installed plugin case passed. The new test uses production-source iOS acquisition/presentation/snapshot assembly and separately built/installed Maestro; native input is intercepted. Derived/missing taps and scrolls refuse before input; reported taps/scrolls succeed. It is not full-packed-core or live-device proof.

Retained failures: initial two missing typed fixture members; wrong check:dead-code invocation (actual check:fallow then passed); first identical-env preflight stopped before GREEN because digests differed; first GREEN feedback passed refusals but failed reported scroll because the fixture expected swipe instead of production's projected gesture. Only the fixture command was corrected, with product bytes unchanged. Parent's final helper has no identical-final-test captured-env RED/GREEN claim. One independent Codex worker now owns that bounded pair and seven ascending exact-head gates. One new Apex MEDIUM review owns the actual final family, including origin and native-create interactions. Both are pending; no running task or CI polling.

Prior committed-clickability hash qualification: completed independent 38e validator recorded eba09af9256ec65f32aada9f4d72b369ac49546cb77a5536428be441f765b3e1, matching committed bytes (message line 87/result assertion 95), with identical captured environment/argv RED 1 → GREEN 0. Earlier parent 9f71/line 63 was pre-format dirty feedback, not a committed Git proof. Old PREP and six-lane exact passes retain their named SHAs and do not transfer to the new family.

Scope: own PREP repair 13 files, +140/−8; cumulative PREP 38 files/794 gross; consumer 104 files/2,467 immediate gross, cumulative 123 files/3,249; DATA 60/762 (64/4,892 without renames), REG 11/377, TOP 74/1,882. Default-main TOP is separately 358 files/9,949 gross. DATA and REG layer patches are byte-identical; TOP added/removed lines match with only index/hunk/Fallow-context differences. Consumer conflict resolution preserves original 502 content plus bounded required assertions. No baseline, threshold, suppression, dependency, export or publication prerequisite was weakened.

No source push, merge, manual CI rerun, device action, npm publication, tag or release occurred. Private-package prerequisites, native ADR acceptance, canonical/native/remote/current-client/live/performance and nightly CodeQL blockers remain open.


Earlier evidence (historical SHA scope)

Latest family: all seven local exact gates passed; source review still pending

The remaining-six validator completed once per lane, with fresh pinned pnpm 11.17 frozen install/build at each reached head. Every receipt has exactCommit=true, checkoutChanged=false, clean initial/final fingerprints, and all skip owners retained.

Candidate Local canonical result Receipt SHA256
Consumer502 affected 42 passed / 25 skipped 2fa72ae1e7af0123ed96a82819177aac8f9d5de4fc19aa03a4e1124d07fe16d3
Consumer502 full 41 passed / 25 skipped ea7296b0716694df2ea74b959ebfbbac5dac995d3693da848b48f6cd3e454ad7
DATA5c5 affected 42 passed / 25 skipped 92be057467572d4b3f1b4cd6ab5603f532860b8d6c832ccb04a37a7530e1645a
REG881 affected 42 passed / 25 skipped e826d33fe391ecc14c6e71f1b5a3309f34a91d18264a3b18ed5ba1cde8124955
TOP9f73 affected 42 passed / 25 skipped a213cef4f9b78a932a4800c7de71cd5a1408e910a4eecd632eee65e8f5f049ba
TOP9f73 full 41 passed / 25 skipped b8234a5a45fe7cffd36502ca50a32a61a51590e1c0b4d3ed6df24a85849758b7

PREP 38e154243c2600d8efccfd951af0b6ad3b9ac7e2 already passed its exact affected gate separately (42/25); it was not repeated. Latest consumer/DATA/REGISTRY/TOP heads remain 502345b69aecbe6aac0bc55bbc7ecb6411e82d4a, 5c5f6b0df310014b951ceb5d1340ec659af0f949, 88128f18cd1e819c5ef618f6b60e984fbf2f1805, and 9f733f36ec1bacc386ba52fec415e57c3e66dcdb. Counts overlap; skips are not passes.

Parent independent verification checked every one of 512 archive files, 511 manifest entries, all 327 original/copy byte pairs (including 308 complete receipt-directory pairs), and all five adjacent hashes. UNSEALED archive SHA256: 10c21368d6d85d544f8b20f98703e2fcdfe09b4d54ceccb37cadd4df52614f60. No existing refs moved or disappeared; only two T3 checkpoint additions were recorded. Default affected base stayed origin/main@d78be6c8, merge-base d38e35e6, with actual 0.22/layer comparisons separate. Lock wait remains unmeasured. Source/checkouts were unchanged; all 19 outer operations terminated successfully.

T3 run metadata: requested 2026-11-14 21:53:14.270Z, started 21:53:15.444Z, completed 23:09:14.868Z (1h15m59.424s elapsed, 1.174s request-to-start). Configured inherited Codex gpt-6.1-sol high/default; executedModel was null, so no independent runtime-identity claim.

The validator reservation is closed. Current reservations are only original A1 and the existing Apex MEDIUM host-port family source review at frozen ce51. A new independent upper native-create recheck is still owed after that review completes. No running delegate was polled, replaced or steered. The source repairs remain unpublished and not source-admitted. Publication waits for the required source reviews; fresh hosted evidence will be required at any rewritten published heads.

Published b057/DDA/d373/f390/2e40 remain unchanged. Their Android failures and DATA's separate native iOS failure remain failed, with historical dispatch causality unproven. No CI rerun, device action, assertion relaxation, or extra local gate occurred. Private placeholder/trusted-publisher/usable publication, canonical/native/live/current-client/ADR/performance and separate nightly security blockers remain open. Pixel7/Pixel9 operating paths stay STOPPED; this does not mean both devices are powered off. No merge clearance.


Earlier evidence before remaining-six completion

Native-create recovery prepared; PREP exact proof verified

New valid finding 4575620206 is separate from clickability: native execution.create() rejects once, but its rejected promise/permanent failure poison later requests. The parent correction clears the failed promise, records recoverable diagnostics separately, clears them on successful next-caller recovery, and retains permanent setup/duplicate failures. TOP keeps typed package/repair/error wrapping.

Latest unpublished upper heads:

  • REGISTRY 88128f18cd1e819c5ef618f6b60e984fbf2f1805 above unchanged DATA 5c5f6b0df310014b951ceb5d1340ec659af0f949: 11 files / 377 gross.
  • TOP 9f733f36ec1bacc386ba52fec415e57c3e66dcdb: 74 files / 1,882 gross. Delta versus earlier clickability TOP ce51 is only two files, +23/−5 (28 gross). Default-main TOP is 357 files / 9,808 gross, disclosed separately.

Parent REGISTRY proof: unchanged test bytes/argv/cwd/captured environment produced RED 1 failed/14 passed, then GREEN 15 passed. The case pins the first failure, later concurrent recovery, cached success, cleared inventory diagnostic, duplicate ownership and disposal. Pinned frozen install/build, full format, typecheck and lint passed. These are local feedback; new upper source review and exact gates remain required. Accepted finding disposition remains open until reviewed correction is published and verified.

The completed clickability validator independently proved captured-environment RED/GREEN and PREP 38e1542 exact affected PASS: 42 passed / 25 skipped, exactCommit=true, checkoutChanged=false. Canonical pack/npm-install/helper execution passed. Receipt SHA256 4023c8083ec939f5e21b0cab6fb9f4223d94d1805468be8e8797984c4674b3e7; verified UNSEALED archive ee01f2586c819f65cd456c38b2e6428bcb704749ef27469704878b20d7e09942, all 161 file bytes and 52 original/copy pairs verified by parent. Receipt UTC 21:35:47.721–21:47:28.171 on November 14. Skipped owners remain obligations, not live/GHA passes.

After PREP passed, an overbroad harness ref-addition guard stopped on the parent-owned new repair alias plus T3 checkpoint additions. No existing ref moved and PREP source/index/checkout were unchanged. The guard stop and earlier preflight assertion incident remain preserved; six later lanes were NOT RUN, not failed or passed. PREP will not be repeated.

A new sole Codex validator now runs exactly those six unrun lanes: consumer502 affected/full, DATA5c5 affected, REG881 affected, TOP9f73 affected/full, with fresh pinned setup at each head and stop-on-first-failure. The existing Apex MEDIUM review retains its frozen ce51 host-port family; a new independent review of 881/9f73 is still owed afterward. Neither worker was polled or steered. All repaired heads remain unpublished; published b057/DDA/d373/f390/2e40 are unchanged.

All four published Android smoke jobs failed at the same Maestro assertion; historical selected-target/coordinate causality remains unproven. DATA's native iOS failure is distinct and unknown. Private npm prerequisites, canonical/native/live/current-client/ADR/performance and separate nightly security blockers remain open. No merge clearance.


Earlier evidence, before native-create recovery and PREP completion

Host clickability repair: unpublished, review and exact validation pending

The completed Android source diagnosis found a P1 inherited by the published preparation and all four consumers: separately bundled private-evidence maps lose the host capture's clickability. The saved hosted failure lacks the failing replay dispatch log, so its selected node/coordinate and causal attribution remain unproven. The completed DATA iOS diagnosis is separate: native .ad back/wait failures remain causally unknown, with no proven DATA regression or matched-base execution.

New frozen local family:

Layer Candidate Actual layer scope
Preparation #4555 38e154243c2600d8efccfd951af0b6ad3b9ac7e2 35 files / 654 gross over root
Consumer #4505 502345b69aecbe6aac0bc55bbc7ecb6411e82d4a 104 / 2,466
DATA #4534 5c5f6b0df310014b951ceb5d1340ec659af0f949 60 / 762; no-renames 64 / 4,892
REGISTRY #4535 d526ac75a6bc491e7478e525d24853366fcee94c 11 / 359
TOP #4537 ce51b133cfea1792f9f9ec84ee85ef964e22e568 74 / 1,882

The required host reader supplies capture-local facts to Maestro's local evidence owner without wire fields, global state or fallback. The regression exercises production source host capture and a separately built/installed plugin, with native input intercepted. It is not full packed-core or live proof.

Parent feedback: intended baseline RED selected inert (20,20) instead of clickable (20,80); unchanged-test GREEN after explicitly rebuilding Maestro passed clickable-first and explicit-index source-order cases. The initial harness setup failure and one stale-plugin run remain failed evidence. RED's complete environment snapshot was not retained; identical captured-environment proof belongs to the new validator. Installed-plugin artifact test, 13 focused tests, format, lint, typecheck, layering and Fallow passed locally. Parent TOP frozen install/build/typecheck also passed. None is a new canonical receipt.

DATA/REGISTRY patches are byte-identical to the published layers. TOP's added/removed lines are identical, with shifted context. Consumer conflict resolution preserves both host contracts; an intermediate parent script error was caught by the endpoint diff, corrected before testing/delegation/publication, and retained in local evidence. Cumulative root-to-consumer is 122 files / 3,108 gross; default-main consumer 271 / 7,217 and TOP 357 / 9,790, including upstream drift.

A fresh Apex MEDIUM source review and sole Codex captured-environment regression/seven-gate ascending validator are running. All five repaired candidates remain unpublished and unadmitted. Existing named-head receipts remain at b057/DDA/d373/f390/2e40; previous source approvals are historical after this P1. No CI rerun, device action, fixture relaxation or gate weakening occurred. Both plugins remain private; npm placeholder/trusted publisher/usable publication, canonical/live/current-client/native/ADR/performance and the separate nightly security blocker remain open. No merge clearance.


Earlier published-family evidence (historical, before the P1 diagnosis)

Final reviewed stack published — 2026-11-14 20:35 UTC

The four replacement heads were pushed together with git push --atomic and explicit per-ref old-SHA leases, immediately T3-linked, then grouped as GitHub stack #4556. The base chain is now 0.22 → #4555 b057 → #4505 DDA → #4534 d373 → #4535 f390 → #4537 2e40. All five PRs are OPEN and ready for review. The former stack retains only merged #4528; no PR or branch was deleted. The first API-only link refused to remove that historical entry, so the documented grouping-only unstack operation preceded creation of the replacement stack.

Current exact local receipts (counts overlap; they are not totals):

Layer/head Gate Passed/skipped Receipt SHA-256
C3 dda310eb578455cb456f45e574952b3c7c639711 affected 42/25 9d40b721ba781ea5d92aa050b76482155947f9e81397fe6dd92994e5c90c4fa7
same DDA full 41/25 16d097745204194bb670ee0b3ac0aee6fe4ce9d7d3f5070b70acec2a469c2ab6
DATA d3739f847fdf560710bddc201fc4a6c7595ae062 affected 42/25 bca69563e85809801d5da6cafc018572e1381869def014d82ddeaff7c55b067e
REGISTRY f39013201319b5f100785580b3e62ccb9ec207a8 affected 42/25 c52e7164730d6aca750a400774b420ae5cba97b876ceb42cfc1b69bf7ad6e812
TOP 2e40c35ee5a777a645cc557a054acb152c854d45 affected 42/25 18f095ff298e7157e54036330306358b70f7e0aea98518c8895df7daf9e4ab1a
same TOP full 41/25 d3cdc266da677781f7a2638ce0fa05cdcf7bd84c55b256e3788a526a8598fc2a

All six raw exits are 0; receipts have exactCommit=true, checkoutChanged=false. Each stage used fresh pinned pnpm 11.17 frozen install/build. TOP's affected ran once 20:10:57.676–20:23:05.174 UTC; full once 20:23:07.198–20:33:40.947. Default origin/main semantics were retained (d78be6c, merge-base d38e35e); actual 0.22 layer scopes were assessed separately. All skip owners remain in receipts. Full mode selects every runnable check; affected retained FAIL-OPEN. Lock wait was not measured.

TOP original/copy files were independently compared (52 affected + 50 full), and all 142 archive files/141 manifest entries verified. UNSEALED archive SHA-256: 774bb2694daee653252ee802cfed38d9f2f7a9899b54d0d9601b81faf816a78a. Parent source hashes, logical index and fingerprint stayed identical; the only shared-ref gain was a T3 checkpoint. The parent returned clean to owned 0.22 at 0081f616a2cfd1088511c4e172db262a08e9529a.

Independent Apex MEDIUM source recheck at exact TOP found no source/design/scope findings and zero new public prose delta. It confirmed removal of the root-relative build helper, restoration of four configs, the unchanged package-local replay guard, and a single validated backend passed through admission. Configured variant is recorded; backend execution effort is unobservable. Review report SHA-256: 00d149710029e7919a8d46f8b6329bba32bbe1c06b6371fef60caeccf1899705. Earlier independent lower-layer and prose verdicts retain their verified unchanged-byte scopes.

All historical failures below remain attributed to their original heads: 411 TS2722, b6cc guidance, 5581 TS18048, eb16 Fallow, 29 R11, parser stop, and experimental no-output/full-byte conditions. The fixture hash difference was only a generated region-path comment; it was not a repository gate. No gate/source/baseline/timeout was weakened and no unchanged gate rerun was used to erase a failure. A parent verifier's full-mode FAIL-OPEN assumption failed, was corrected to check full selection, and passed without rerunning validation or changing evidence.

Readiness stays separate: rewritten-head hosted checks are now pending native events, not inherited from old heads. Prep b057's existing hosted 19-success/2-skipped, actual Coverage 8/8 proof remains b057-only. Both plugins remain private. Maintainer npm placeholder, release-workflow trusted publisher and usable publication are required before any plugin/prep merge. Native ADR acceptance, canonical Pager/Nav ARM, representative native/HTTP recording/platform/remote/current-client/live verification and performance remain open. Pixel7/Pixel9 routes stay stopped. No merge, release, npm publication, tag or workflow dispatch occurred.

The older running/pending descriptions below are retained historical snapshots and are superseded by this exact-head publication record.


Ascending exact-head validation stopped at REGISTRY, with no retries or source edits by the validator.

Head / invocation Raw result
dda310eb578455cb456f45e574952b3c7c639711 affected 0; 42 passed / 25 skipped
Same DDA full 0; 41 passed / 25 skipped
d3739f847fdf560710bddc201fc4a6c7595ae062 affected 0; 42 passed / 25 skipped
5581b74a491ef70a22911efd32fc66e7c2c7bbe2 affected 2; 2 passed / 25 skipped / 1 failed / 39 not run
7a4a0de8c1dc7de386aed636ab893fe2da461730 affected/full NOT RUN

REGISTRY failed TS18048 at packages/plugins/src/registry.ts:309 and packages/maestro/src/plugin.test.ts:83–84: setup may provide native execution instead of a format backend. All four receipts have exactCommit=true, checkoutChanged=false; pinned pnpm 11.17.0 setup/build passed at each executed head. Counts overlap. Lower-stage replay directories remained absent.

Parent independently verified all 458 archived file bytes and 158 original/copy pairs. Receipt SHA256s, in table order:

9d40b721ba781ea5d92aa050b76482155947f9e81397fe6dd92994e5c90c4fa7
16d097745204194bb670ee0b3ac0aee6fe4ce9d7d3f5070b70acec2a469c2ab6
bca69563e85809801d5da6cafc018572e1381869def014d82ddeaff7c55b067e
74461b6dc601f709b4e3465418572732a1e3b5ef4fa6d49b20ed2232eb612cba

Archive SHA256: 9094ef9d5a58017ef1f88161c391540decc0755dc239ab25e1651dfb7f48b8cd (verified, unsealed).

The subsequent repair at REGISTRY f39013201319b5f100785580b3e62ccb9ec207a8 passed its own affected gate: raw 0, 42 passed / 25 skipped. Its bounded regression was RED 1 failed/16 passed (expected invalid_plugin vs plugin_load_failed) → identical GREEN 17 passed. Exact clean receipt SHA256: c52e7164730d6aca750a400774b420ae5cba97b876ceb42cfc1b69bf7ad6e812.

TOP eb165f460d4e1bc11bfec0abbccd3189e97cfd76 then FAILED affected at Fallow: raw 1, 5 passed / 1 failed / 25 skipped / 36 not run; full was NOT RUN. Exact clean receipt SHA256: 1fbc7e9e1283daee1a906212b69620b163d74b9610d8cec77a07185c6bc41e2f. Parent verified all 145 archived file bytes and 59 original/copy pairs; verified-unsealed archive SHA256: 03deab9d342942f4dded43ac672da33358be04838558eeaacfbe53edd1ef6b43.

Fallow JSON attributes zero introduced dead-code findings (8 inherited), one introduced complexity finding in replay admission and two introduced build-config clone groups. Existing baseline exclusions are unchanged. The original failure remains failed; no gate was retried by the worker.

New local TOP 29bbb35f0fe1ff4c62bece88b5fd744bebf83e66 separates suite and entry admission and shares the existing workspace-bundling/fail-unresolved-import policy across five configs. Entries, declaration settings, dependency overrides and package privacy remain explicit. Its immediate routing scope is 79 files/1,918 gross lines; cleanup delta is 7 files, +58/−62. Parent dirty feedback: typecheck, Fallow, 11 existing router tests and build passed. The intermediate type-import and partial-Fallow cleanup failures remain recorded; no baseline, budget, suppression, exclusion or gate change was used.

The first 29bbb35 validator stopped before canonical gates: real tsdown threw UNRESOLVED_IMPORT but left a 123-byte partial output. The parent-supplied no-output condition remains FAILED; that harness checked rejection only and returned 0. Its 89-file archive is verified-unsealed (a231900d7e4223aa626ba63736065e8266c4c35fa41fda85832d241969ed0775).

The subsequent matched old-inline/new-shared policy comparison passed its bounded condition: both real build subprocesses exited 1 with the same diagnostic and identical partial output SHA256 9e6bfe20a8d136ee3b1c448b0db72b842f87236d3437b619811d4d90cb2c560b. Neither promises transactional zero output. Checked-in build/pack chains and publish job dependencies require build success. No product cleanup or repository oracle was weakened.

29bbb35's first affected gate then FAILED R11: 3 passed / 1 failed / 25 skipped / 38 not run, exactCommit=true, checkoutChanged=false. Five package configs imported the root-level helper outside their package boundaries. Full validation was NOT RUN; Fallow was not reached. Receipt SHA256: cfcee0badacc8e3d51a09e22aa83ed899881d046f7b6c8c804683fc8cc550bda.

Parent verified all 225 archived files, 2 adjacent report/verifier files, 89 prior evidence original/copy pairs and 5 gate original/copy pairs. The first inventory check omitted a nested prior manifest and failed; that record and archive are unchanged. Independent verification established the complete archive bytes. Archive SHA256: fa6b93e1025e1d549b00515514c1767417c3ac10d2380e9314d58a61a9f81fa0 (verified, unsealed).

The first a1256cc validator stopped after successful install/build on a build-log parser error, before probe or gates. A verified parser then captured 18 Maestro and 5 replay outputs; the actual local guard rejected UNRESOLVED_IMPORT with Node exit 1. Its full-byte output condition failed at 171 versus 123 bytes. Parent independently verified all 256 archived files (archive 887f566f96ff60da37e4cbfc1017ff763fd8b0fe164b8685e97accf485bc8bf8) and compared the saved outputs: only line 2's generated source-region path comment differs; import, executable statement and export are byte-identical. Both experimental conditions remain recorded as failed; no repository gate was weakened, and no further probe was run.

The completed independent review of 29bbb35 confirms the blocking R11 finding and recommends the smaller per-package design. It also suggests passing the validated backend to admission helpers rather than re-reading flags. Parent accepted both: local candidate 2e40c35ee5a777a645cc557a054acb152c854d45 removes the root helper, restores four configs to eb16, keeps replay's guard package-local, and passes one validated backend value through both helpers. No new exports/dependencies, R11 exceptions, baselines or test relaxations. The replay config is byte-identical to the actual a125 probe producer.

Final scope is 74 files/1,882 gross over unchanged REGISTRY f390; versus a125 only admission changes (+11/−6). Parent dirty typecheck and all 11 router cases passed. A NEW independent Apex medium source recheck owns the final candidate. Parent is now running its first real repository affected/full sequence at the exact committed head, with pinned fresh setup and default base semantics. No canonical command at a125 was run or repeated.

The current affected run has passed typecheck, layering, Fallow and package stages; its overall result remains pending. Full runs only if affected passes. Source review and local validation remain separate. All public consumer/replay heads stay unchanged until required review/validation clear. Hosted/live/private publication prerequisites, native ADR and performance remain blocked; no merge clearance.

@thymikee
thymikee force-pushed the t3/maestro-plugin-extraction branch from 0158fac to dda310e Compare November 14, 2026 20:34
@thymikee
thymikee removed this pull request from stack #4529 November 14, 2026 20:35
@thymikee
thymikee changed the base branch from 0.22 to fix/c3-prep-atomic-layer-20261114 November 14, 2026 20:35
@thymikee
thymikee added this pull request to stack #4556 November 14, 2026 20:35
@thymikee

Copy link
Copy Markdown
Member Author

At dda310e, the workflow path, fuzz close() and registry retry items from the earlier review are fixed. The plugin boundary still has one defect that fails CI, plus an install hint and missing live evidence.

The packed plugin bundles its own copy of snapshot-private-evidence.ts (tsdown.config.ts#L21 uses alwaysBundle). That copy has its own module-level WeakMap (line 42). The daemon writes clickability and viewport evidence into its own instance, and the plugin reads from the bundled one (runtime-clickability.ts#L48), so the lookup always misses. On Android, a no-index tapOn on duplicates then taps the inert first duplicate. This matches the Android Smoke failure in run 43638215079: resolve duplicate target through no-index Maestro tapOn → assertVisible 'Maestro selection: clickable' did not match. The base #4555 runs pass the same job. On iOS, assertSnapshotScreenOrigin finds no viewport evidence, so it stops refusing coordinate taps when the window origin is unknown. Unit tests share one module instance, so they cannot see this. Should every capture-private fact the daemon writes and the engine reads reach the engine through ReplayBackendHost, as the delta already does for emitDiagnostic and registerSensitiveValue? One host-owned evidence accessor, implemented by the daemon with its own contracts instance and called by the engine's readers, would remove the dependence on a shared singleton. Projecting the evidence onto the snapshot data that invoke hands the plugin would also work. I did not confirm whether invoke passes the same snapshot object. A small gate that lists module-level WeakMap or registry state in the packages alwaysBundle pulls in could keep this from returning. Please re-run Android Smoke at the new head.

The help maestro text (replay-backend-declarations.ts#L44) and the plugin-not-installed error (replay-backend-registry.ts#L93-L102) still tell users to run agent-device plugins add @agent-device/maestro. packages/maestro/package.json is still private: true at 0.0.0, so npm cannot install it, and plugins.md and run-maestro-flows.md now say so. Users follow the CLI hint, not the website caveat. Every surface that names the install command should name a package the release can install. Either make the package publishable in this layer (drop private, set a real version, add a changeset) and remove the "private" doc sentences, or hold this PR until publication lands.

This change is device-facing: tap dispatch, animation and screenshot waits, runScript children and the diagnostics sinks now run from a separately bundled plugin. No live packed-plugin run is attached, and the Android lane that exercises the changed route is failing. After the evidence fix, please install the pnpm pack tarball with agent-device plugins add <tarball> and start a daemon from the new head. On one Android and one iOS device, run replay flow.yaml --maestro --debug on a flow that reaches a no-index selector tapOn on duplicates (for example the maestro-clickable-first fixture) and a waitForAnimationToEnd or screenshot wait. Please paste the success output, the plugins list --json line with the store-installed version, the maestro_tap_dispatch and maestro_screenshot_compare diagnostic lines, and the Android timing against the ADR 0015 baseline. A green Android Smoke run at that head is also needed.

The Android Smoke failure looks related to this PR: the store-installed plugin route is new in this layer, the base #4555 and unrelated PRs pass, and only this stack fails that step. Apple passed. I did not run builds, tests or the packed smoke, and I inferred the mechanism from the bundle config and the failing step. The private-package install thread still applies. These threads are fixed at this head and can be resolved: android.yml workspace path, apple.yml workspace path, fuzz close memoised, test cleanup try/finally and replay failure retries.

Before merge, route the daemon-owned snapshot evidence to the plugin so Android Smoke passes, attach the live Android and iOS plugin runs, and reconcile the install hint with the private package.

Land the consumer migration, isolated test installation and resolver setup, artifact retirement, dependency removal and matching enforcement as one self-contained layer. Preserve the existing plugin spine rank instead of lowering its boundary.
@thymikee
thymikee force-pushed the t3/maestro-plugin-extraction branch from dda310e to 924a0bd Compare November 15, 2026 01:14

@apex-sentinel-staging apex-sentinel-staging Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Apex Sentinel: 1 new finding on 924a0bd. The verdict and the full list are in the summary comment.

Comment on lines +57 to +75
export async function openReplayBackend(id: ReplayBackendId, env: NodeJS.ProcessEnv) {
const { AppError } = await import('@agent-device/kernel/errors');
const [{ storePluginCandidates }, { resolveProviderPlugins, startProviderRegistry }] =
await Promise.all([
import('@agent-device/plugins/store'),
import('@agent-device/plugins/registry'),
]);
const registry = await startProviderRegistry(
await resolveProviderPlugins(await storePluginCandidates(env)),
env,
{
adaptWebDriver: async () => {
throw new AppError('INVALID_ARGS', 'Client replay scope cannot start a provider');
},
},
'replay-backends',
);
try {
const { createReplayBackendHost } = await import('./replay-backend-host.ts');

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

🆕 🟡 Minor · Client replay operations now evaluate every installed provider plugin package

A client with TestMu/webOS/Tizen installed running agent-device replay --maestro now evaluates each of those packages; if one throws or performs I/O during module evaluation, the Maestro run fails or slows. openReplayBackend() calls resolveProviderPlugins over all store plugin candidates instead of filtering to only the replay-backend package it needs.

Suggested change

In openReplayBackend, narrow storePluginCandidates(env) to candidates whose package equals getReplayBackendDeclaration(id).package before resolveProviderPlugins, so a client-side replay scope loads only the backend package it needs.

Details

Also raised here

Also raised at this location by persona javascript: Client replay scope eagerly imports every installed provider plugin — Only setup is skipped by the new 'replay-backends' scope; module evaluation and its side effects still happen on a path whose only job is engine lookup. This also means a provider package that throws at import time makes the replay/export command fail, where previously it degraded to a failed inventory entry. Cause: backend selection reuses the provider resolution path, which loads all candidates to read their identity.

persona: senior-developer, javascript

Useful? Reply @apex-sentinel-staging like, or @apex-sentinel-staging dislike <why> if it is not.

Reply @apex-sentinel-staging note ... to add context, or @apex-sentinel-staging pivot ... if it is wrong — I will re-check the cited code.

Copy link
Copy Markdown
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Accepted for unnecessary module/factory evaluation in the client-side scope. storePluginCandidates creates lazy loaders, but resolveProviderPlugins evaluates every loader before the replay-only setup scope takes effect. A provider import error itself is caught into inventory, so I do not attribute a replay-command failure to that throw; unrelated evaluation, side effects and blocking I/O are the concrete issue.

Local C3 9025146e239005dac2b838f9edf7e21d40d92ff5 resolves the vetted backend package once and filters store candidates by that package before registry loading. Metadata discovery still reads the store; this does not claim zero unrelated metadata I/O. Embedded resolvers and the daemon's complete registry path remain unchanged. The existing absence regression now installs an unrelated package whose evaluation writes a marker and throws; it retains the typed missing-plugin/install-command assertions and requires the marker to stay absent.

The correction is retained through local TOP 66d0daa23ccac8f92167bf40b3b2d04b269dfcf8, preserving the package-command and documentation fixes. Full repository format passed; new test execution, intended RED→GREEN, independent review and exact-head gates remain pending behind the current serialized validator. No runtime pass or speedup is claimed. Keeping this thread open until reviewed, validated publication.

@thymikee

Copy link
Copy Markdown
Member Author

Thanks for the update. I reviewed 924a0bd against the earlier findings. The host-routed evidence read replaces the bundled WeakMap lookup that missed at dda310e, so the Android no-index tap failure should be fixed. Two findings remain.

The install command still names a package that npm cannot serve. help maestro and the plugin-not-installed error (message, installCommand and hint in replay-backend-registry.ts) tell users to run agent-device plugins add @agent-device/maestro. The declaration is unchanged since dda310e, and packages/maestro/package.json is still private: true at 0.0.0. A user who runs replay --maestro on a release from this layer will follow the hint and fail at npm. Every surface that names the install command (the help body, the error, plugins.md, run-maestro-flows.md and the changeset) must name a package the shipping release can install. Can you make @agent-device/maestro publishable in this layer (drop private, set a real version, add a changeset) and delete the "unavailable" doc sentences? Otherwise please hold the merge until publication lands.

The change is device-facing, and no live run covers it yet. Tap dispatch, the screen-origin refusal, clickability ordering, the animation and screenshot waits, and runScript children now run from a separately bundled plugin across ReplayBackendHost. The new packed tests use a fake invoke boundary, as the PR body says. Lint, Typecheck & Package (which runs them) and the Apple smoke lanes were cancelled at this head, and I could not confirm a green Android Smoke. So the fix for the earlier Android Smoke failure is not proven, and I see no defect from the cancellations themselves. Please run pnpm pack in packages/maestro, then agent-device plugins add <tarball>, then start a daemon from 924a0bd. On one Android and one iOS device, run replay flow.yaml --maestro --debug on a flow with a no-index tapOn on duplicates (the maestro-clickable-first fixture) and a waitForAnimationToEnd or screenshot wait. Please paste the success output, the plugins list --json line showing the store-installed version, the maestro_tap_dispatch and maestro_screenshot_compare diagnostic lines, and the Android timing against the ADR 0015 baseline. Please also link green Android Smoke and Lint, Typecheck & Package runs at this head.

The open inline thread on narrowing provider candidates applies only in part. Evaluating every installed provider module is a minor, optional speedup. The claim that an import-time throw fails the replay is wrong, because loadSlot catches it into slot.failure, so please resolve that part.

There are no conflicts. Before merge, the install command and package must agree, and the live Android and iOS runs and green checks above must be attached.

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant