Decide Microsoft 365 approvals in the chat and tidy orchestration retries

[Paul Lizer (paullizer)]

Summary

• Microsoft 365 approvals in a plan are decided in the chat. A SharePoint or OneDrive plan step that needs more than a quick read (3 downloads, 25 MB or 12,000 tokens of file content) stops for the user's extended-analysis approval. Before, it only linked to Approvals. Now the stopped message explains why and offers Allow this time, Always allow for SharePoint/OneDrive or Quick read only. A choice is saved through the existing approvals API, and the plan continues. The failure now carries the approval id; only application-minted ids are kept.
• No "Retry this failed step?" for Microsoft 365 stops. A Use an action step that stopped for Microsoft 365 sign-in, approval or policy can only have read data, so the server no longer requires external-effect confirmation for it (recovery_projection / prepare_retry). Agent steps and other failures still ask. A confirmation that is still needed now closes as soon as the retry is admitted, instead of staying open until the whole retry finishes.
• A retry's answer replaces the stopped attempt. The stopped attempt's "The request could not be completed." message is hidden once a later attempt exists, including after reload. It is left out of model history (orchestration and normal chat) and exports, but stays stored. Successful retries no longer show "Saved execution attempt - Attempt 2"; Message details has View previous attempt / Review saved attempt.
• Why the approval appeared at all: it was the extended-analysis guard with the default Ask preference, not access consent. That's documented in the fix doc, along with the existing per-source preference under Settings.

Linked issue

None; no issue was filed for this work. The companion change that cites Microsoft 365 files, emails and calendar events like workspace documents is developed separately on paullizer-m365-source-citations.

Release Notes & Latest Features

• New Feature
• Bug Fix
• UI Enhancement
• Breaking Change
• Internal only

Is this visible to end users?

• Yes
• No

Is this admin-facing (Admin Settings, governance, deployment, config)?

• Yes
• No

Should this become a Latest Feature card?

• Yes
• No
• Already added

Screenshot needed for the card?

• Yes
• No
• Attached

Version bump

• application/single_app/config.py VERSION third segment bumped, or not needed because this is docs-only (0.261.301 → 0.261.302)
• deployers/version.txt bumped, or not needed because deployers/ was not changed

Testing / validation

Run with Python 3.12 in an isolated venv whose only difference from the shared interpreter is pyOpenSSL 26.4.0. The shared interpreter's pyOpenSSL 25.3.0 doesn't import with cryptography 50.

• python -m pytest functional_tests/test_orchestration_m365_approval_retry_flow.py -q: 8 passed (new).
• python -m pytest ui_tests/test_v2_orchestration_m365_inline_approval.py ui_tests/test_v2_orchestration_recovery.py ui_tests/test_v2_orchestration_m365_recovery.py ui_tests/test_v2_orchestration_outputs.py -q: 115 passed (7 new inline-approval tests).
• python -m pytest ui_tests/test_v2_orchestration_recovery_backend.py -q (with PYTHONPATH=application/single_app;functional_tests): 18 passed. The reload test was updated: the stopped attempt is now hidden after a retry.
• python -m pytest on test_analyze_backend_saved_integration, test_chat_stream_history_context_fix, test_image_chat_route_integration, test_workflow_result_chat_routes, test_saved_results_container_access_chat_orchestration, test_content_screening_history, test_workflow_chat_delivery_placement_and_masking, test_orchestration_conversation_context, test_orchestration_m365_actions and test_orchestration_recovery_safety: 172 passed, 9 failed. The same 9 fail identically on a clean checkout of the base commit (test_chat_stream_history_context_fix ×4, test_workflow_result_chat_routes ×1, test_workflow_chat_delivery_placement_and_masking ×4), so they are not caused by this PR. Three hand-built test namespaces gained the new history helper.
• python -m pytest functional_tests/test_orchestration_*.py --continue-on-collection-errors (with PYTHONPATH): 6,916 passed, 69 failed. None of the 69 are caused by this PR:
  • 55 also fail on a clean checkout of the base commit fa3422dee. Rerunning their 6 files gives the identical 62-test failure set on both checkouts; 53 of those are ImportError: cannot import name 'enabled_required' from 'functions_settings' in a test stub.
  • The other 14 are test_orchestration_failure_telemetry cases that fail only in the combined run (test-order pollution). That file passes alone on both checkouts (18/18).
• npm run typecheck and npm run build -- --outDir ../../ui_tests/artifacts/orchestration-plan-editor in application/v2_ui: passed.
• python scripts/check_xss_sinks.py --base-sha fa3422dee --head-sha HEAD <changed files>: passed for 13 files.
• python scripts/check_broken_access_control.py --base-sha fa3422dee --head-sha HEAD <changed files>: passed for 7 files.
• python scripts/build_docs_inventory.py: no surface change. python functional_tests/test_docs_app_surface_coverage.py: 7/7. python functional_tests/test_docs_site_quality.py: 6/6.

Documentation

• Release notes updated, or not needed (v0.261.302)
• Feature documentation updated, or not needed (not needed)
• Fix documentation updated, or not needed (docs/explanation/fixes/ORCHESTRATION_M365_APPROVAL_RETRY_FLOW_FIX.md; docs/admin/orchestration.md and docs/reference/chat-controls.md updated)

Security checklist

• New Flask routes include @swagger_route(security=get_auth_security()) (no routes added)
• Settings sent to non-admin frontends use sanitize_settings_for_user() (no settings sent)
• Browser JavaScript is served from local SimpleChat static assets only; no CDN-hosted JS
• No secrets, keys, connection strings, or local-only artifacts are included