Skip to content

chore: release mobilerun 0.6.19 - #446

Merged
RasulOs merged 1 commit into
mainfrom
agent/release-0.6.19
Sep 10, 2026
Merged

chore: release mobilerun 0.6.19#446
RasulOs merged 1 commit into
mainfrom
agent/release-0.6.19

Conversation

@RasulOs

@RasulOs RasulOs commented Sep 10, 2026

Copy link
Copy Markdown
Collaborator

Prepare Mobilerun 0.6.19 with the fixes merged in #442#445. Align the root package, compatibility-shim version, and all five shim dependency pins at 0.6.19; update only the lockfile's own-package version. All locked dependency versions are preserved.

Portal compatibility was merged in droidrun/gists#1: exact mapping 0.6.190.7.25, preserving every existing mapping. The public installed-package resolver returned Portal 0.7.25 after CDN propagation.

Included release changes

Validation

Python 3.13.9, macOS ARM64. Source changes are limited to three versioning files; no README, agent-test-flows, credentials, fixture, or evidence files are included.

  • Guide unittest discovery: 89 tests passed (0.356s).
  • Locked full suite: 978 passed + 3 subtests (24.13s).
  • Fresh runtime dependencies, installed release wheel, repository-locked pytest 9.0.3: 978 passed + 3 subtests (18.78s).
  • Focused tracing/provider/CLI matrix (test_langfuse_v4, test_grok_api, test_phoenix_tracing, test_provider_defaults, test_openai_oauth_llm, test_grok_cli):
OpenTelemetry SDK/API OpenAI SDK Result Duration
1.40.0 2.30.0 175 passed 2.98s
1.40.0 2.54.0 175 passed 38.95s
1.44.0 2.30.0 175 passed 38.04s
1.44.0 2.54.0 175 passed 40.22s

Langfuse was 4.15.1 in all environments. Fresh runtime resolution also used Anthropic 1.5.0, google-genai 2.22.0, mobilerun-sdk 5.4.0, mobilerun-core-local 0.6.0, llama-index 0.14.23, and llama-index-llms-google-genai 0.11.1. Exact complete manifests are retained outside the repositories.

  • Black 26.5.1: 185 files unchanged. Ruff checks on Python files changed since v0.6.18 passed. git diff --check passed in both repositories.
  • Built wheels and source distributions for both packages. Twine checks passed; the shim retains pre-existing missing-long-description warnings.
  • Clean installed-wheel environments verified site-packages imports, package versions, public APIs, all six defaults, and mobilerun/droidrun CLI entry points. Installing the built droidrun[anthropic] wheel against the built Mobilerun wheel resolved correctly; all five shipped shim pins are exact. Dependency checks passed in all seven environments.
  • Real installed Portal resolver, with the proposed map supplied locally: 0.6.19 resolves to 0.7.25; 0.6.8 and 0.6.18 still resolve to 0.7.22.

Android end-to-end results

Restarted Medium_Phone_API_36.1, verified Android 16 and Portal 0.7.25. Ten sequential real MobileAgent sessions used the installed release wheel with freshly resolved runtime dependencies, saved credentials, implicit provider defaults, indexed tapping, a 15-step limit and a 180-second agent timeout. Destinations and handler markers were checked independently from the model's completion message; selected indices, coordinates, raw trees, screenshots and action logs are retained externally.

Provider / auth Implicit model Scenario Steps Duration Result
Gemini / OAUTH gemini-3.8-flash-tiered Portal app information 6 52.03s PASS
Gemini / OAUTH gemini-3.8-flash-tiered Settings search → Display & touch 4 30.12s PASS
Gemini / OAUTH gemini-3.8-flash-tiered Decorative parent, clickable child, overlapping targets 5 30.90s PASS
Gemini / OAUTH gemini-3.8-flash-tiered Stealth tapping; focusable and non-focusable popups 6 31.54s PASS
Gemini / OAUTH gemini-3.8-flash-tiered Blocked target, error feedback, dismissal and retry 4 27.69s PASS
Gemini / API gemini-3.8-flash Settings search → Display & touch 4 29.86s PASS
Anthropic / API claude-sonnet-5 Settings search → Display & touch 4 28.97s PASS
Anthropic / OAUTH claude-sonnet-5 Settings search → Display & touch 4 27.04s PASS
Openai / API gpt-6-astra Settings search → Display & touch 4 31.36s PASS
Openai / OAUTH gpt-6-astra Settings search → Display & touch 4 41.05s PASS

Gemini OAuth Settings search used the real Langfuse SDK with local export capture and mocked upload I/O: 36 unique exported spans, 5 screenshot spans, 5 valid media references and exactly 5 upload jobs. Original spans remained unchanged; one owned Langfuse pipeline, no duplicate exports and no normalization errors were observed. This verifies SDK transformation and queuing, not a hosted Langfuse dashboard or external upload service.

Installed-wheel CLI checks passed: device discovery (1.20s), ping (2.14s), and the guide's screen-report smoke with a two-step limit (one step, 10.25s). The agent reported Settings and called only complete(success=True); independent before/after trees confirmed the same Settings screen.

Test setup observations

The first fully fresh test run with pytest 9.1.1 produced 977 passes and one warning-count assertion failure. A standalone reproduction with no Mobilerun imports proved that pytest attaches the same capture handler twice when a logger switches propagation, capturing one LogRecord twice. Rerunning with the repository-locked pytest 9.0.3 passed the full suite; every freshly resolved runtime package was identical. No tests were skipped or weakened and no dependency pins were changed.

The initial CLI smoke configuration omitted _version and was correctly rejected before agent startup. The external harness was corrected to use the installed config format version and rerun. This was a test setup error.

Publication and post-publication checks are complete; results follow.

Publication and post-publication verification

Released as https://github.com/droidrun/mobilerun/releases/tag/v0.6.19 at merge commit 0837a7ddbd537d91f818d614a0f9ada20e4c6a6e with an annotated tag.

  • Publishing workflow https://github.com/droidrun/mobilerun/actions/runs/34512527715 passed all four jobs: version check, both-package build, TestPyPI, and PyPI.
  • Both packages have 0.6.19 wheels and source distributions on TestPyPI and PyPI. All four published artifacts have identical SHA-256 hashes to the locally validated builds.
  • Fresh public-PyPI installations of mobilerun==0.6.19, droidrun==0.6.19, and droidrun[anthropic]==0.6.19 passed installed import/API/CLI/version checks and dependency checks. All five shim pins are 0.6.19; the Anthropic extra resolves correctly.
  • Each public installation resolves Mobilerun 0.6.19 to Portal 0.7.25 using the live public mapping.
  • Final Android 16 smoke from the clean PyPI-installed Mobilerun package passed: device discovery 1.76s, ping 2.17s, and Settings screen-report task 10.46s / one step within the two-step limit. Only complete(success=True) ran, and independent before/after trees confirmed Settings.

Package pages: https://pypi.org/project/mobilerun/0.6.19/ and https://pypi.org/project/droidrun/0.6.19/.

@chatgpt-codex-connector

chatgpt-codex-connector Bot commented Sep 10, 2026

Copy link
Copy Markdown

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review Status Commit Review trigger
📝 Code Review Completed 2026-09-10T18:07:49.740774Z 1df8ae5 PR opened
ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

  • Open a pull request for review
  • Mark a draft as ready
  • Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

@RasulOs
RasulOs merged commit 0837a7d into main Sep 10, 2026
6 checks passed
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