0.5.0.dev0 development source and retained packages#
The current source is 0.5.0.dev0 development work. The supervised Qwen27 work
and standalone documentation were published on codex/selective-local-execution;
the repository presentation cleanup prepares that work for integration into
main. Check the actual branch and commit before treating that integration as
complete. See the publication qualification addendum
and complete workflow.
Consolidating development source into main is distinct from publishing a
release or recording a passed human/live acceptance result. Keep the eleven-row human ledger, Strata
standalone blockers and separate OpenHands/Strata qualification visible. The
released v0.4.1 tag/package and installed controller remain separate from main.
The earlier companion-release-candidate provenance below is historical evidence,
not the current activation scope.
The companion-release-candidate branch is development work, not a qualified Windows release candidate. Its historical branch name grants no acceptance. Source publication does not merge into main, create a release/tag, distribute a binary, send a callback or pass a human test.
Source provenance#
The historical publication base is f6df420181f21856cb421619e2c1310f6cf46685
(then main and the existing v0.4.1 tag target). The canonical repository is
nishdel/OmniLAB-Local-Agent-Bridge-;
the former LocalAgentBridge URL redirects there.
The existing unpublished history was reviewed before source publication and retained without rewriting:
| Commit | Included work |
|---|---|
cc88d4d86d5d038513d694c542a00339093495b4 |
Preserved v0.5 workflow and protected worker baseline |
53a2c59a1a1a95d29de0ed4f528c9da02c7c7e9b |
Durable companion jobs, exact models and managed reviews |
8b02eba3deb02cc24393421b317644b183114a09 |
UI cleanup and portable lifecycle fixtures |
91ee3a20bb869c36e23bb1c7813827de053637b3 |
Integrated workflows, recovery repair, tests and qualification records |
The integration commit has tree 43266f6c38b3cfa4edf99a7953d5d0b2c71db33e.
Its runtime, tests, examples and profiles equal qualified source tree
3a7d24cbf8c8d41c14d88fa6a6df326a6fa37ebb; only the Windows qualification
addendum was added after that qualified tree. Subsequent manual-testing,
navigation, public-manifest and provenance changes are documentation-only and
belong to a separate commit. Use Git history to identify that documentation
commit; it is not the retained package's source snapshot.
Unchanged retained artifacts#
| Artifact | SHA-256 |
|---|---|
| localagentbridge-0.5.0.dev0-py3-none-any.whl | 3fde801aa34b5769ed68185ed7dfd857825d819f3af807f49bcc798ce37bbb2c |
| localagentbridge-0.5.0.dev0.tar.gz | 7afe70772012d09328d4cb38617a06a982fb10173cc2def9f2bab3efd7208f10 |
These artifacts predate the documentation commit. They were neither rebuilt from this newer documentation snapshot nor uploaded with the source branch. The public manifest now includes the later qualification/manual documentation; that does not retroactively change either archive or its earlier export.
Validation and outstanding acceptance#
Retained results describe the qualified runtime: Windows 411 passed/1 skipped, installed recovery 50 passed, and actual Python 3.11.9 installed-package checks 361 passed/1 deselected (the accepted recovery suite was excluded). See recovery and qualification for scope. Recovery covers the tested process-crash model, not host power loss. Unsaved output and ambiguous incidents still require explicit intervention. Local completion never implies supervisor approval or Codex verification.
Source-push checks are separate: Ruff, documentation links/generated reference, the public-source/ignore audit and four release-audit unit tests. No runtime change or new package qualification is claimed by these checks. Pattern scans and manual review reduce disclosure risk but cannot prove absence of every possible secret. Private evidence, personal configuration, transcripts, runtime databases, models, virtual environments and build artifacts remain local.
Remaining release gates are human DPI/readability/keyboard/focus/hide/reopen and status observations, fresh Windows installation, the privilege-dependent symlink check, owner-defined performance criteria and scored observations, and separately authorized live callback testing. Authentication/quota fixtures do not prove live behavior. The existing private callback approval request must be reviewed before any authenticated invocation; source-push authorization does not authorize it. No existing model preference or inventory changes are needed.
Use the manual definitions and issue-body draft. Their new rows are No / Not tested. Execution results belong in one future issue, not this source report. Opening that issue requires separate authorization.