release: promote verified develop snapshot 97f6425484 to main#15637
Merged
Conversation
Co-authored-by: Shaw <shawgotbags@gmail.com>
Update the docker CI smoke crash scanner so optional plugin resolver skip lines do not fail agent image builds while real top-level package crashes still surface.
Adds mobile-build-smoke build timing budget gates for iOS simulator, iOS full-Bun simulator, and Android cloud-debug builds.
…he 07-04 pin bump (#15575) Restores the llama.cpp fork pin that includes Kokoro G2P/IPA exports while preserving Gemma4 conversion support.
…he RAM floor (#15576) Allows the LP3 native agent runtime to bypass the local-agent RAM floor while keeping local-model RAM limits intact.
…d registry app-core runs vitest with `isolate: false` (maxWorkers:1), so every test file shares one module registry. Five suites install module-level mocks of shared modules and re-import them into that registry via `vi.resetModules()` for their own self-defense, but none clear the registry on teardown. Under isolate:false those mocked modules persist in the cache after the file finishes and leak into whichever real-module suite the sequencer happens to run next: - `ensure-route-min-role` / `database-rows` install a full-replacement `@elizaos/shared` mock (no `...actual`), stripping the real `normalizeDeploymentTargetConfig` / `normalizeServiceRoutingConfig` normalizers — leaking into `first-run-persistence.restart`. - `auth-pairing-routes` mocks `./auth/sessions` (stubbed `createMachineSession`) and `../services/auth-store` — leaking into `auth-pairing-flow`, whose real AuthStore mint is then a no-op so `findSession` reads back null. The victim is non-deterministic because vitest's size-based sequencer order varies run to run (observed as `auth-pairing-flow:293` and `first-run-persistence.restart` on different runs of the same tree). Fix at the correct layer: each mock-heavy suite clears the shared module registry on teardown (`afterAll(() => vi.resetModules())`) so its mocks cannot outlive the file. No assertion, status code, or mock behavior is changed. Full app-core lane is green 3/3 consecutive runs (164 files, 0 fail), each modified suite passes standalone, and all five polluters + both former victims pass in one combined run. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
test(app-core): stop mock-heavy suites leaking mocks across the isolate:false registry
Removes the stale mobile null-stub for @elizaos/plugin-aosp-local-inference so Android all-local bundles expose ensureAospLocalInferenceHandlers.
…ico) Regenerated from packages/app/public/brand/favicons/app-icon-512.png with the macOS squircle mask; replaces the stale gold Milady art the desktop bundle was shipping. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…kflow's artifacts build-android.yml died at gradle :app:copyForkLlamaLib — the task hard-requires the fused libelizainference.so set for arm64-v8a, and CI had no staging step (no ~/.eliza, no in-workflow NDK cross-compile). The fused libs are built by build-llama-ffi-android (green again since #15565), so stage its latest successful artifacts and export ELIZA_MTP_ANDROID_LIBDIR{,_X86_64} for gradle. Fails loud with a pointer at the ffi workflow when no green run exists. Fixes the build-android half of #15540
The staged fused lib exports the fused-voice ABI, so the elizavoice JNI wrapper compiles against the omnivoice headers inside the plugins/plugin-local-inference/native/llama.cpp submodule — which the workflow never initialized (layer 2 of #15540).
…us-4-8 Both the SDK session and the CLI path hard-defaulted to claude-opus-4-7, a generation behind the current Opus. Deployments that set ELIZA_CLI_CLAUDE_MODEL were unaffected; anyone on the bare default was silently pinned to the older model. Test pins updated; 94/94 green.
…n-device agent
On a sub-12GB device the on-device agent has no local text model
(DeviceRamTierPolicy), so in hybrid mode its @elizaos/plugin-elizacloud must
reach Eliza Cloud (→ Cerebras) for TEXT_LARGE/TEXT_EMBEDDING. The agent runs as
a separate Bun process and ElizaAgentService passed it NO cloud credential, so
plugin-elizacloud booted unauthenticated ('No provider registered for
TEXT_LARGE') and the agent could not generate.
Read the account's Eliza Cloud API key from the CapacitorStorage pref
'eliza:cloud-inference-api-key' (+ optional base URL) and inject it into the
agent process env as ELIZAOS_CLOUD_API_KEY / ELIZAOS_CLOUD_ENABLED. No-op when
the pref is absent (pure-local / not-yet-onboarded), so existing flows are
unchanged.
Proven on a real Light Phone III (6GB, with #15576): pref set → agent restart →
'[CloudAuth] API key validated' → the on-device agent answered 'Say hello in
exactly three words' with 'hello there, friend.' via Cerebras, and its planner
selected the TAKE_PHOTO device action for a camera request. The renderer half —
writing the key to the pref during hybrid onboarding — is the paired follow-up.
Refs #14390, #15539
Contributor
|
Claude encountered an error —— View job I'll analyze this and get back to you. |
Bypass queued CI per operator instruction.
Contributor
LifeOps Multi-Tier BenchmarkSuite:
|
Bypass queued/unstable CI per operator instruction.
Contributor
LifeOps Multi-Tier BenchmarkSuite:
|
Contributor
|
Claude encountered an error —— View job I'll analyze this and get back to you. |
Bypass queued/unstable CI per operator instruction.
Bypass queued/unstable CI per operator instruction.
Bypass queued/unstable CI per operator instruction.
Bypass queued/unstable CI per operator instruction.
Contributor
|
Claude encountered an error —— View job I'll analyze this and get back to you. |
Contributor
LifeOps Multi-Tier BenchmarkSuite:
|
Contributor
|
Claude encountered an error —— View job I'll analyze this and get back to you. |
Bypass queued/unstable CI per operator instruction.
Bypass queued/unstable CI per operator instruction.
Contributor
|
Claude encountered an error —— View job I'll analyze this and get back to you. |
Member
|
Promotion branch refresh / local verification update (2026-07-09) Merged latest Local checks run on the updated head: Visual/OCR/color evidence was also run locally against the generated screenshots via: Summary: Manual screenshot review:
The GitHub checks for the new head are still queued/running. I will inspect any completed failures once Actions reports them. |
Contributor
LifeOps Multi-Tier BenchmarkSuite:
|
Contributor
LifeOps Multi-Tier BenchmarkSuite:
|
Contributor
LifeOps Benchmark —
|
Contributor
LifeOps Benchmark —
|
NubsCarson
moved this from In progress
to Done
in Launch QA — Eliza app + Cloud (#13406)
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Promotes verified develop snapshot
97f6425484to main → prod. Closes the 82-commit gap (prod was on9ff78abfb).Carries (highlights)
Migrations (2, additive — migrate-db gate will run + needs reviewer approval)
0173_agent_backup_verification.sql0174_agent_backup_latest_index.sqlBoth additive (backup verification + index); no destructive/omnibus schema change.
Gate
Standard promote: migrate-db (prod-env reviewer) + full CI. Do not merge until the meaningful lanes are green and migrate-db is approved. QA-gated by [qa-agent]; sign-in matrix (Google/Magic/Passkey/Discord/GitHub) verified live on prod, EVM fix verified on develop.
Evidence