[Kelly (lykelly19)]

No description provided.

[Copilot]

🟡 Changes recommended

Stop requests can leave reminders enabled because overdue timers are excluded from the stop path and concurrent turns allow a later stop to overtake an earlier scheduling request.

3 open issues

• Interpretation before locking lets later turns overtake earlier ones · New
• Overdue pending timers become undiscoverable to stop requests · New
• Scheduling requests scan all routines twice · New

What changed in this PR

Adds a Routines Autopilot sample that turns personal Teams messages into timers and recurring reminders using the Foundry Routines API.

Changes:

• Adds model-based request interpretation, routine lifecycle management, status queries, and reminder delivery.
• Provides deployment configuration and provisioning scripts to create or reuse Azure resources.
• Documents setup, publication, safety limits, and operational verification.

File	Description
samples/​python/​autopilot-agents/​routines/​scripts/​prepare-resources.ps1	Selects and validates Azure resources and model deployments.
samples/​python/​autopilot-agents/​routines/​scripts/​prepare-provision.ps1	Invokes the sample’s provisioning helper.
samples/​python/​autopilot-agents/​routines/​requirements.txt	Pins hosted runtime dependencies.
samples/​python/​autopilot-agents/​routines/​README.md	Documents behavior, deployment, safety, and troubleshooting.
samples/​python/​autopilot-agents/​routines/​pyproject.toml	Defines package dependencies and the command entry point.
samples/​python/​autopilot-agents/​routines/​main.py	Provides the hosted runtime entry point.
samples/​python/​autopilot-agents/​routines/​infra/​modules/​foundry.bicep	Creates new Foundry resources while referencing reused resources.
samples/​python/​autopilot-agents/​routines/​infra/​main.parameters.json	Maps environment settings to infrastructure parameters.
samples/​python/​autopilot-agents/​routines/​infra/​main.bicep	Coordinates resource provisioning and deployment outputs.
samples/​python/​autopilot-agents/​routines/​azure.yaml	Configures hosting, publication, environment settings, and hooks.
samples/​python/​autopilot-agents/​routines/​agent/​routines.py	Implements scoped ownership checks and routine lifecycle operations.
samples/​python/​autopilot-agents/​routines/​agent/​app.py	Interprets requests and handles scheduling, status, stopping, and delivery.
samples/​python/​autopilot-agents/​routines/​agent/​activity_routing.py	Routes supported Teams messages and routine delivery events.
samples/​python/​autopilot-agents/​routines/​agent/​__main__.py	Enables module-based startup.
samples/​python/​autopilot-agents/​routines/​agent/​__init__.py	Initializes the agent package.
samples/​python/​autopilot-agents/​routines/​.gitignore	Excludes local environments, caches, and configuration.
samples/​python/​autopilot-agents/​routines/​.agentignore	Excludes development and provisioning files from deployment.
samples/​python/​autopilot-agents/​README.md	Adds Routines to the Autopilot sample catalog.

_{Review effort: Balanced}

[Copilot]

🔵 Needs a closer look

The routine lifecycle lacks automated safety tests, and reminder send failures are swallowed rather than surfaced for failure reporting or reconciliation.

1 open issue

• Overdue pending timers become undiscoverable to stop requests

2 previously missed issues

In code that hasn't changed since last review

Surface unconfirmed routine deliveries without blindly retrying

samples/​python/​autopilot-agents/​routines/​agent/​app.py:435

Routine-delivery events also use this helper, so a send timeout or transport error is swallowed and the delivery handler returns normally without confirmed delivery. Separate best-effort conversational replies from reminder delivery, and surface unconfirmed deliveries through the host's failure-reporting or reconciliation path. The application error handler must preserve that failure without sending a generic chat reply or blindly retrying an ambiguous send.

Add lifecycle tests for malformed records and ambiguous commit outcomes

samples/​python/​autopilot-agents/​routines/​agent/​routines.py:339

The new routine lifecycle has no automated tests, although comparable lifecycle safety is tested in meeting-assistant/tests/test_meeting_lifecycle.py. Add offline tests for foreign or malformed records, multiple active routines, re-enabling an identical disabled routine, and create/enable/disable operations that commit but time out. Assert outcomes and mutation counts for matching, missing, and conflicting readbacks so reconciliation cannot silently repeat a mutation or report an unconfirmed success.

2 resolved since last review

• Interpretation before locking lets later turns overtake earlier ones
• Scheduling requests scan all routines twice

_{Review effort: Balanced}