Release v3.0.0 · Azure/agent-app-ui

Agent Landing Zone dual-read and rebrand (Azure/agent-landing-zone#695). App
Configuration now loads the agent-lz label, which overrides the legacy
gpt-rag label; AGENTLZ_* keys are read before legacy GPT_RAG_* keys.
Tracer names use the agentlz. prefix. Visible branding (assistant name,
welcome title, release footer) now reads "Agent Landing Zone". Existing
gpt-rag deployments keep working unchanged.

Retry failed classic-chat uploads without losing attachments. Restore the
previous session conversation ID when a failed upload allocated a new one.
Reattached files are ingested under a fresh conversation before the question
runs; existing conversation ownership checks remain unchanged.
Stop questions after unconfirmed ingestion. False results and raised
failures finalize a reference-bearing retry notice without sending the
question or claiming successful file processing. Only confirmed batches
update uploaded-document bookkeeping.
Keep OpenAPI failures retryable and truthful. Failed schema generation
propagates instead of returning a successful metadata-only schema; later
requests retry generation and cache only success.
Expose owner-index repair needs without weakening panel access. A failed
optional index write logs that repair is required while the turn continues;
missing ownership rows still hide panel listings and deny read, feedback,
and delete operations.

Preserve dependency failures without masking implementation defects.
Narrow configuration, HTTP/JSON, Blob, JWT, logging and VERSION handlers to
their concrete failure contracts. Fix the unbound retry notification and
avoid retrying absent in-memory settings. Record retained application
boundaries individually as unapproved proposals with failure-test evidence.
Separate Chainlit history adaptation from history operations. Put the
data-layer callbacks, factory and consume-once session bridge in api.history;
keep history authorization, orchestrator operations and the single user cache
in services.history with explicit context. Preserve legacy exports and
ownership-before-session-selection ordering without a storage rewrite.
Canonical UI runtime package with legacy adapters. Move runtime ownership
into src/gpt_rag_ui, share citation rendering between chat and restored
history, and separate callback registration from reusable operations. Keep
uvicorn main:app, existing configuration/defaults and externally staged
Chainlit assets; install the package in the existing container build.

Isolate protected quality-tool execution. Run static tools without
candidate module shadowing or candidate package/build-backend installation
in the evaluator; analyze the explicit source graph in an isolated worker.
Add executable shadowing and installation-isolation regression fixtures.
Proposed protected-base Python quality gates. Add pinned development-only
Ruff, mypy and import tooling, individual typing debt records, full static
dependency checks, an explicit broad-handler inventory and a fail-closed
quality-gate aggregate alongside unittest. Initial handler approvals,
bootstrap review and administrator activation remain outstanding; this
change does not claim required merge enforcement is active.
Exact quality policy and executed failure evidence. Validate closed
governance schemas, stable move/debt identities, annotation/suppression sites,
adapter exports and dynamic imports. Bind exception approvals to distinct
protected operations and same-run, non-skipped unittest results. Require
protected aggregation and ephemeral offline Linux-image acceptance without
publishing an image or changing repository settings.
Close binding and namespace quality-check bypasses. Detect qualified
suppression decorators on untyped functions, invalidate exception records
when effective catch bindings change, and reject unresolved catches.
Discover root namespace modules and retain their full typing/import identity
without changing application imports or exception behavior.

Navigation Menu