oh-my-openagent v4.19.4

LAST RELEASE BEFORE THE OMO NATIVE CLI PUBLIC RELEASE

This is the final release in the current product line before the OmO Native CLI becomes public. The next public release begins the native CLI era.

Smarter model routing across every harness

Model configuration now uses one canonical reasoning vocabulary from shared config through OpenCode and Codex. Updated fallback chains route Mini workloads through Luna and DeepSeek options, improve Kimi-based category routing, and expose targeted warning controls when a configured category is unavailable.

More reliable recovery during long-running work

Runtime fallback now recognizes retryable usage-limit status events through shared patterns, preventing recoverable sessions from being missed. Goal-driven work also treats an armed monitor, background child, or other live resumption channel as progress instead of incorrectly declaring the run blocked.

Stronger research and session tooling

The shared skill set now includes the data-scientist workflow, an expanded research delivery process with steering excursions and closing briefings, and browser-agent session discovery through the Aside history scanner. Planning activation is also more deliberate: the planning workflow starts only when the user explicitly requests it.

Cross-platform and daemon hardening

Windows path handling is more robust in audit jobs, Codex preserves executable permissions for the bundled codegraph runtime, and daemon ownership checks now use authenticated ping paths before reaping a process. Several asynchronous tests were also made event-driven to remove timing-dependent failures.

---

• b072d2791 Merge pull request #6542 from code-yeongyu/release/v4.19.4-source-state
• 89bd6dd18 Merge pull request #6541 from code-yeongyu/docs/release-guidance-native-cli
• 1d689fe64 docs(evidence): record release guidance QA
• 2acc5f54e test(release): capture failing release notes proof
• 0ddc2b5e8 test(release): add release notes validator
• 03121f8d2 docs(release): exclude contained adapter changes
• bcc1b3161 docs(publish): enforce Native CLI release notes
• d16c27b94 Merge pull request #6540 from code-yeongyu/fix/runtime-fallback-status-pattern-drift
• 6fbf93542 fix(runtime-fallback): share retry patterns for status events
• 7cbcc53a7 Merge pull request #6539 from code-yeongyu/feat/plan-gate-hardening
• 7eb2f788a docs(skills): activate ulw-plan only on explicit user request
• 532a6ca6d docs(evidence): record PR 6521 CI green
• 22ab05de7 Merge pull request #6526 from code-yeongyu/fix/codegraph-windows-lock-drift
• b010c1f4f Merge remote-tracking branch 'origin/dev' into fix/codegraph-windows-lock-drift
• fb5c1c12c fix(ci): use native path in GPT Mini audit
• 84e8c4d1b Merge pull request #6524 from code-yeongyu/fix/gpt-mini-audit-windows-path
• f4b1b0f93 fix(ci): resolve audit root path on Windows
• f60ab7529 Merge pull request #6518 from code-yeongyu/feat/luna-deepseek-chains
• f9b535c7e fix(omo-codex): preserve codegraph executable modes
• b3b45cd2a feat(model-core): migrate Mini chains to Luna and DeepSeek
• 888a26b61 Merge pull request #6514 from code-yeongyu/fix/ci-codex-git-bash-hooks-timeout
• 10b839953 test(omo-codex): isolate git bash hook fixture
• d4e7b1c42 Merge pull request #6512 from code-yeongyu/feat/ulw-research-excursions-and-closing-briefing
• c7fc22369 feat(skills): add steering excursions and a closing briefing to ulw-research
• 252922ca1 Merge pull request #6511 from code-yeongyu/feat/category-fallback-chain-tuning
• 96e7a028c feat(model-core): retune category fallback chains
• 33435840a fix(skills-loader-core): align the debugging builtin description with the trimmed shared source
• 37d4c624e Merge pull request #6508 from code-yeongyu/feat/vendor-data-scientist-skill
• 6097b72ab test(omo-codex): classify the vendored data-scientist quick-query script
• b4af75550 feat(skills): vendor the data-scientist skill into shared-skills
• df2a90efe Merge pull request #6507 from code-yeongyu/feat/shared-ulw-research-delivery-workflow
• c05008977 Merge pull request #6506 from code-yeongyu/chore/skills-remove-contract-test-guidance
• afdf13819 feat(skills): port the ulw-research delivery workflow to the shared variant
• 8b8eb7120 chore(skills): drop contract-test guidance from the programming skill
• f18648e1a docs(agents): mandate the change-execution protocol for user-ordered patches
• 4b9c1724f test(model-routing): align fallback-chain fixtures with the gpt-5.6-luna-fast swap
• bc9295823 Merge pull request #6501 from code-yeongyu/refactor/unified-task-status-target
• 25ae49fa0 feat(model-routing): swap gpt-5.4-mini-fast to gpt-5.6-luna-fast(low)
• 65b788bab Merge pull request #6494 from code-yeongyu/fix/auto-update-checker-network-test
• 64524483f test(auto-update-checker): mock npm registry fetch in getLatestVersion tests
• f2872273d Merge pull request #6491 from code-yeongyu/fix/task-reject-category-with-model
• 6895e89c3 Merge pull request #6483 from code-yeongyu/feat/ulw-goal-footer-animation
• 06d1077c8 Merge remote-tracking branch 'origin/dev' into feat/ulw-goal-footer-animation
• 258fab041 Merge pull request #6479 from code-yeongyu/fix/cleanup-session-end-async-flake
• 317a24578 test(agent-env): use FIFO completion signal
• 36d8d1a39 Merge remote-tracking branch 'origin/dev' into fix/cleanup-session-end-async-flake
• 0eac4b018 test(agent-env): eliminate async cleanup timing race
• 3236be40f Merge pull request #6478 from code-yeongyu/fix/task-child-fallback-isolation
• 25905bcba Merge pull request #6476 from code-yeongyu/fix/codegraph-sessionstart-windows-flake
• 9049f1847 test(codegraph): stub process sweep in SessionStart tests
• 5c253bad7 test(ci): record daemon owner attestation evidence
• 845f36514 fix(utils): use real daemon ping and auth paths
• 96909d08b fix(utils): bind daemon reap to owner identity
• 9e91e889e Merge remote-tracking branch 'origin/dev' into HEAD
• 915c5e4a9 Merge pull request #6473 from code-yeongyu/fix/ultrawork-goal-block-monitor-wait
• ca4ffad91 Merge pull request #6472 from code-yeongyu/feature/aside-agent-sessions-scanner
• 5b97ecc11 fix(ulw): treat waiting on a live resumption channel as progress, never blocked
• 82f0efbed test(omo-codex): register aside_scanner.py in the Python migration inventory
• cb8d15136 docs(shared-skills): document the Aside session store and linkage
• aaaba202b test(shared-skills): cover Aside scanner parsing and platform registration
• f2a1d761b feat(shared-skills): add Aside browser-agent session scanner
• 487e5bb2f test(ci): record macOS flake verification
• 6c959aeda test(codegraph): signal descendant readiness before timeout
• 5110f5479 test(lsp-core): stabilize diagnostics fallback timing
• 0aadead8d Merge pull request #6455 from code-yeongyu/feat/reasoning-config-unification
• dbcdab9d7 Merge remote-tracking branch 'origin/dev' into feat/reasoning-config-unification
• 047f0889b feat(omo-opencode): consume canonical reasoning through delegation end to end
• 77135c427 fix(omo-opencode): lower direct-agent reasoning instead of assigning variant blindly
• 65b85d658 feat(omo-codex): thread the canonical reasoning field through the Codex installer
• deef131f7 fix(config-migration): report the actual winner when reasoning keys conflict
• 046afd07b Merge remote-tracking branch 'origin/dev' into feat/reasoning-config-unification
• 04750e45b refactor(omo-config-core): keep the reasoning vocabulary inside the config core
• 260f219c3 fix(schema): generate omo and legacy schema artifacts from their own builders
• 17d486764 fix(omo-config-core): stop treating the account home .omo as a project layer
• aa320d846 Merge remote-tracking branch 'origin/dev' into pr-6451-review
• 0118768e2 fix(codex): regenerate installer dist for the 4.19.3 version sync
• 37563bb7e feat(omo-config-core): canonicalize model tuning on a unified reasoning field
• d2ccd7127 feat(omo-opencode): lower the canonical reasoning field to OpenCode model settings
• e7bd8a143 feat(omo-codex): source Codex model_reasoning_effort from the unified reasoning field
• 83009aedf feat(model-core): add unified reasoning-level vocabulary with colon model suffix
• 82854b4cc Keep default model routing current across harnesses (#6447)
• ba8e41f45 Keep default model routing current across harnesses
• 5decafdf7 fix(qa): reject unsafe process tree PIDs
• a6cd227b6 Merge pull request #6446 from code-yeongyu/feat/runtime-chain
• efb3d152c test(opencode): make shared skills package check deterministic
• f1d8a2987 Merge pull request #6443 from code-yeongyu/docs/chain
• 5ca5a3f45 docs: remove apitopia and document chain overhaul
• 1ec03deb5 Merge pull request #6441 from code-yeongyu/feat/dead-chain-warnings
• 7c6abe231 test(opencode): quote Kimi provider fixture key
• ace13396a Merge pull request #6439 from code-yeongyu/feat/fallback-architect-tip-and-give-me-tips
• 91ddee749 fix(config): align category routing contracts
• c73d1102a fix(omo-opencode): quote kimi-for-coding key in delegate-task test fixture
• dabb9cbfb test(omo-opencode): align expectations with the new category chains
• 0bb6f1d3e fix(omo-opencode): mirror warn_unavailable category key from omo-config-core
• 98fb15fb8 Merge remote-tracking branch 'origin/dev' into feat/fallback-architect-tip-and-give-me-tips
• bdbdc63fa feat(omo-opencode): port new category defaults and plumb the deep requiresModel gate
• 9897bce97 Merge pull request #6434 from code-yeongyu/feat/model-core-chains
• e93a8534e feat(model-core): replace apitopia rungs with kimi-for-coding and overhaul category chains
• 5aa9bbc81 Merge pull request #6433 from code-yeongyu/feat/warning-schema
• ab825471f feat(omo-config-core): add unavailable-category warning suppression keys
• 511e776ce Merge pull request #6127 from vvvzhangshuo/fix/sisyphus-junior-gpt-variant-override
• 17241fa88 test(sisyphus-junior): record real-opencode QA evidence for GPT override fix
• 9b47ef23a fix(sisyphus-junior): apply variant and reasoningEffort overrides independently
• e03a16e3c Merge remote-tracking branch 'origin/dev' into pr-6127-work
• 0a43f7922 fix(sisyphus-junior): respect GPT variant and reasoningEffort overrides

Thank you to 4 community contributors:

• @vvvzhangshuo:
  • fix(sisyphus-junior): respect GPT variant and reasoningEffort overrides
• @etehofk1-ops:
  • fix(qa): reject unsafe process tree PIDs
• @MoerAI:
  • Keep default model routing current across harnesses
  • Keep default model routing current across harnesses (#6447)
• @minpeter:
  • test(lsp-core): stabilize diagnostics fallback timing
  • test(codegraph): signal descendant readiness before timeout
  • test(ci): record macOS flake verification
  • Merge remote-tracking branch 'origin/dev' into HEAD
  • fix(utils): bind daemon reap to owner identity
  • fix(utils): use real daemon ping and auth paths
  • test(ci): record daemon owner attestation evidence
  • test(codegraph): stub process sweep in SessionStart tests
  • Merge pull request #6476 from code-yeongyu/fix/codegraph-sessionstart-windows-flake