Testing Chat Changes

Disclaimer

This specific page uses temporary drafts written and maintained by Generative AI. While verified to be accurate, please cross-verify with source code.

How to use the chat regression tests, which pin the decisions made before a provider is called: whether the bot replies (shouldBotReply() in src/utils/chat/replyDecision.ts) and which personas answer, in what order (determineMatchingPersonas() in src/utils/chat/triggerProcessor.ts).

bun test tests/regression/chat/

The tests need no Discord token, API key, or database. They build mock Message, Client, and TextChannel objects with only the fields these functions read.

Fixtures

Inputs are in tests/regression/chat/fixtures/conversations.json and expected results in expected-decisions.json beside it. The fixtures cover a direct mention, deliberate @trigger mode and its rejection of plain trigger words, an autochat channel assigned to an alter, multi-persona ordering, and a persona webhook that must not trigger itself.

To add one:

1. Add the conversation to conversations.json with stable fake IDs (user_004, channel_001).
2. Add its expected decision to expected-decisions.json.
3. Run bun test tests/regression/chat/.

Each fixture protects one behavior. When you extract logic out of tomoriChat.ts, add fixtures that call the new module directly. When you fix a chat bug, add its fixture in the same change.

Behaviors worth a fixture when a chat stage changes: tool-call routing (REST and MCP), multi-persona order, webhook self-trigger suppression, empty mentions and malformed reply references, rate-limit and cooldown rejection, and provider fallback or missing credentials.

chat.regression.test.ts has a skipped [REGRESSION PROBE] that inverts one expectation. Remove its .skip, confirm it fails, and restore it to check that the tests still detect a wrong decision.