Previously seeds carried hand-curated {category, target, initialHint}.
Now SEEDS is a flat string[] of keywords — at round-start, the model
generates {category, initialHint} on the fly. Benefits:
- adding a seed is trivial (just append a word)
- every round gets a fresh cryptic opener (varies across plays of the
same word)
- HINT STYLE rules apply to the opening hint too, so the initial clue
isn't a definitional giveaway
Implementation:
- prompts.buildStartRoundPrompt(target) — with good/bad examples
- ai-client.generateRoundStart(env, target) — same JSON-in-content
approach as judge(), with defensive fallbacks + redactSecret
- handlers.startFreshGame now async; surfaces roundstart errors via the
existing UPSTREAM_FAIL path
Tests: 449 pass (5 new for generateRoundStart, 1 for roundstart error path).
Gemma 4 likely rejects the flat "traditional" tools schema we were sending
(the docs use OpenAI-wrapped shape for this model) — causing env.AI.run to
throw and users to see the "AI service hiccup" reply every turn.
Switch to the universal approach:
- system prompt asks the model for a one-line JSON {is_guess, answer, hint}
- ai-client.extractText handles both Workers-AI and OpenAI response shapes
- parseJudgementJson walks brace-depth to extract JSON from stray prose /
accidental code fences
- logs twentyq_ai_throw / twentyq_ai_unparseable with preview on failure
so future issues surface in wrangler tail immediately
Tests: 7 new (parser + extractText); 444 total pass.
- SqlStore interface + CF D1 wrapper + per-module factory (table prefix convention)
- init signature extended to ({ db, sql, env }); sql is null when DB binding absent
- custom migration runner walks src/modules/*/migrations/*.sql, tracks applied in _migrations table
- npm run db:migrate with --dry-run and --local flags; chained into deploy
- fake-d1 test helper with subset of SQL semantics for retention and history tests
grammY-based bot with a module plugin system loaded from the MODULES env
var. Three command visibility levels (public/protected/private) share a
unified command namespace with conflict detection at registry build.
- 4 initial modules (util, wordle, loldle, misc); util fully implemented,
others are stubs proving the plugin system end-to-end
- util: /info (chat/thread/sender ids) + /help (pure renderer over the
registry, HTML parse mode, escapes user-influenced strings)
- KVStore interface with CFKVStore and a per-module prefixing factory;
getJSON/putJSON convenience helpers; other backends drop in via one file
- Webhook at POST /webhook with secret-token validation via grammY's
webhookCallback; no admin HTTP surface
- Post-deploy register script (npm run deploy = wrangler deploy && node
--env-file=.env.deploy scripts/register.js) for setWebhook and
setMyCommands; --dry-run flag for preview
- 56 vitest unit tests across 7 suites covering registry, db wrapper,
dispatcher, help renderer, validators, and HTML escaper
- biome for lint + format; phased implementation plan under plans/
tiennm99