umputun / tg-spam / 29291997932

84%

master: 84%

Build Type

Pull #427

github

Committed by umputun

Commit Message

fix(review): scan authored text in prohibited-langs and unify settings validation

The prohibited-language check ran on req.Msg, which bot.OnMessage builds by
concatenating the user's text with quoted/reply-to content. A user replying to a
prohibited-script message could reach the per-script threshold from text they did
not author and be permanently banned, with no LLM veto. Add
spamcheck.Request.AuthoredMsg and AuthoredText(); OnMessage records the user's own
text and Check passes AuthoredText() to the hard block. Softer content checks still
see the full concatenated Msg.

Also address two pre-existing issues surfaced by the review:
- move all cross-field settings invariants into config.Settings.Validate(), shared
  by startup, save-config, and the web save handler, so an invalid combination
  entered via the web UI is rejected before it is persisted rather than bricking
  the next --confdb startup.
- use t.TempDir() for the sqlite storage suite instead of a shared $TMPDIR path
  that raced with other packages under "go test ./..." and produced SQLITE_BUSY.

Add the missing paranoid/approved-user scope test and refresh the CLAUDE.md
prohibited-languages and quoted-text sections.

Pull Request Pull Request #427: feat: prohibited languages spam detector

Coverage Stats

131 of 136 new or added lines in 8 files covered. (96.32%)

185 existing lines in 3 files now uncovered.

8819 of 10540 relevant lines covered (83.67%)

234.16 hits per line