Skip to content

feat: add mobile harness strategy modes - #7

Open
Harzva wants to merge 124 commits into
mainfrom
main-dev
Open

feat: add mobile harness strategy modes#7
Harzva wants to merge 124 commits into
mainfrom
main-dev

Conversation

@Harzva

@Harzva Harzva commented Jun 20, 2026

Copy link
Copy Markdown
Owner

Summary

This PR promotes the Mobile Harness strategy layer to release-readiness for non-counted app-side dry traces.

It is non-counted strategy release-readiness, not a formal benchmark result. All dry trace outputs remain gated with counts_as_experiment=false / non-counted status, and real provider/tool/device callback promotion remains blocked until explicit evidence requirements are met.

What Changed

  • Added the strategy ablation schema, registry, scaffold/pilot run artifacts, validators, and documentation.
  • Added app-side strategy services and runners:
    • StrategyDispatcher
    • strategy_runners.dart
    • memory packet bridge via memory_packet_service.dart
  • Added StrategyModeCard to the Home Tools page for:
    • Auto
    • ReAct
    • Plan-Execute-Verify
    • Supervisor/Handoff
    • Experimental Swarm
  • Added Android emulator QA evidence for the strategy UI dry trace flow.

Android Emulator QA

Android emulator QA passed for the latest release APK.

  • Evidence directory: docs/mobile-harness-benchmark/strategy-ablation/evidence/android-strategy-qa-20260620/
  • APK SHA256 is recorded in apk-sha256.txt.
  • install.txt records Success.
  • launch.txt records Status: ok for com.mobilecode.app/.MainActivity.
  • Details screenshots/XML are included for Auto, ReAct, PEV, Supervisor/Handoff, and Experimental Swarm.
  • App-scoped logcat evidence contains no FATAL EXCEPTION, E/flutter, ANR, or MissingPluginException.

CI / Local Verification

  • Mobile Runtime CI: success
  • flutter test test/services/ test/widgets/strategy_mode_card_test.dart: 307 passed
  • flutter analyze lib/widgets/strategy_mode_card.dart: No issues found
  • python3 -m py_compile scripts/run_mobile_harness_strategy_ablation.py scripts/validate_mobile_harness_strategy_ablation.py scripts/generate_mobile_harness_strategy_callback_pilot.py: passed
  • Strategy validators for r1-scaffold and p4c-callback-pilot: passed
  • git diff --check: passed
  • High-confidence secret/private-path scan: clean
  • README/GitHub Pages link check: 16/16 returned 200

Review Focus

Please focus review on:

  • mobile_agent/lib/services/strategy_dispatcher.dart
  • mobile_agent/lib/services/strategy_runners.dart
  • mobile_agent/lib/services/memory_packet_service.dart
  • mobile_agent/lib/widgets/strategy_mode_card.dart
  • docs/mobile-harness-benchmark/strategy-ablation/release-readiness-2026-06-20.md
  • docs/mobile-harness-benchmark/strategy-ablation/evidence/android-strategy-qa-20260620/

Merge Checklist

Before merge, confirm:

  • PR checks are green.
  • No secrets, private paths, raw transcripts, cookies, or .env values are included.
  • counts_as_experiment=false and non-counted strategy boundaries are preserved.
  • Experimental Swarm remains feature-gated and reports experimental_strategy_disabled unless explicitly enabled in a later phase.

After Merge

After merge, release/tag or GitHub Pages/README link updates should be handled as a separate release step.

Not Included

The local worktree still contains unrelated dirty/untracked work that is intentionally not included here, including local model runtime work, architecture assets, zip review packs, and snake workspace files.

Harzva added 30 commits June 20, 2026 06:52
Add non-counted strategy dispatcher/runners, memory packet bridge, strategy mode UI, strategy ablation docs, validators, and Android emulator QA evidence.
P6.0-P6.2 runtime benchmark contract scaffold
Harzva added 30 commits August 7, 2026 05:28
security: harden public mobile release credentials
feat: activate verified MobileCore Omni runtime
…ction-link

feat: fail closed on MobileCore background restriction
release: MobileCode v0.1.78 background-safe MobileCore linkage
ci: distinguish System UI ANR from MobileCode failure
docs: record v0.1.78 iOS build evidence
test: make physical MobileCore QA fail closed
docs: record fail-closed dual-app regression
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