VahterBanBot: ephemeral warning on spam auto-deletion (flag off by default)

[Szer]

What/why

A newcomer whose message gets auto-deleted as spam currently gets zero feedback — they don't understand what happened, repost, and after ~4 deletions the karma auto-ban (CheckAndAutoBan) bans them in all monitored chats. This adds a best-effort ephemeral Bot API 10.2 message (visible only to that user, in the same chat), sent right after a qualifying auto-deletion, warning them not to repost. Text is deliberately fixed/generic (no scores, no ML/LLM distinction, no strike counts) so it can't be used to reverse-engineer detection. Off by default (SPAM_WARNING_ENABLED=false).

Gating rules (Bot.fs's DeleteSpam, locked design)

Warning is sent only when ALL of:

1. SpamWarningEnabled is true (read via IOptions<BotConfiguration>.Value at call time, never captured).
2. CheckAndAutoBan returned false for this deletion (the user was NOT just total-banned — a ban notice, not a warning, is the right signal there).
3. reason is MlSpam, LlmSpam, or ContentFilterSpam. Explicitly excluded: InvisibleMention, SpamTextCacheHit, ReactionSpam — these are near-zero-false-positive spammer signatures, and warning there would only educate spammers.

Uses CallIgnore (never CallExn) — a delivery failure never fails the deletion, matching the existing confirm//vahter_report ephemeral patterns.

Already-banned path finding

The "already banned" fast-delete path (BotService.JustMessage, ~Bot.fs:1401) has its own inline deletion code — it does not call DeleteSpam at all. So it's structurally excluded from ever producing a warning; no additional guard was needed there. Locked in as a regression test (SpamWarningAlreadyBannedTests).

Enable SQL (hand-run only — settings are never Flyway-seeded)

INSERT INTO bot_setting (key, value) VALUES ('SPAM_WARNING_ENABLED', 'true')
ON CONFLICT (key) DO UPDATE SET value = EXCLUDED.value;

After running, POST /reload-settings to pick it up without a redeploy. SPAM_WARNING_TEXT is also bot_setting-backed (defaults to the bilingual RU/EN text baked into Program.fs) if the wording ever needs tuning live.

Tests (tests/VahterBanBot.Tests/SpamWarningTests.fs, all green)

1. Flag ON + ML spam deletion of a non-banned user → exactly one ephemeral sendMessage (receiver_user_id = sender, chat_id = chat, warning text).
2. Flag OFF (default) + same scenario → no warning.
3. Flag ON + deletion that triggers the karma auto-ban (CheckAndAutoBan → true) → no warning.
4. Flag ON + SpamTextCacheHit deletion (ban-seeded cache, enforce mode) → no warning.
5. Flag ON + late message from an already-banned user (JustMessage's fast-delete path) → no warning.

dotnet test tests/VahterBanBot.Tests -c Release: 258 passed, 0 failed (includes the 5 new tests above, individually re-verified green as well). dotnet build clean for both touched projects.

🤖 Generated with Claude Code

Co-Authored-By: Claude Fable 5 noreply@anthropic.com
Claude-Session: https://claude.ai/code/session_01Wi7gKmshkHfVtSB3tA4gmJ