[AliZakaee]

Summary

• distinguish partial Telegram auto-reply delivery from an unconfirmed send
• alert the owner with the customer and delivery outcome so they can inspect the chat before retrying
• preserve the invariant that outbound history and conversation memory are written only after complete delivery
• add regression coverage for both failure modes

Why this is meaningful

Automatic replies can exceed Telegram's single-message limit and be delivered in chunks. The send adapter already reports when a later chunk fails, but the auto-reply pipeline previously allowed that exception to reach a log-only wrapper. A customer could receive a truncated reply while the owner saw no alert. Complete send failures were similarly invisible outside logs. This change makes those system-level outcomes actionable without changing successful sends, approval mode, storage schemas, or public interfaces.

Validation

• Baseline: .venv/bin/pytest — passed, 276 tests
• Targeted: .venv/bin/pytest tests/unit/handlers/test_business_updates.py -q — passed, 18 tests
• CI-equivalent: .venv/bin/pytest --cov=. --cov-report=term-missing --cov-report=xml --cov-report=html --junitxml=junit.xml — passed, 278 tests; 60% total coverage
• Syntax: .venv/bin/python -m compileall -q config.py database.py handlers services telegram_api.py utils.py tests — passed
• Diff hygiene: git diff --check — passed
• Secret/forbidden-path scan — passed; only the handler and its unit tests changed

The local Python 3.9 runtime emitted urllib3's existing LibreSSL compatibility warning; it did not affect the passing suite. Remote CI covers the supported Python 3.10, 3.11, and 3.12 matrix.

Risk and rollback

Risk is low: failure paths now send one additional owner alert, while the success path is unchanged. Roll back by reverting the feature commit.

Prepared with automated assistance by EQ for AliZakaee.