Gilyumei eizou#1
Conversation
|
@codex すべての問題、バグ、致命的なバグ、ミス、バグの匂いを指摘しろ。これは映像認識のバリアント改良と、識別処理の向上である |
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: ccd5a920b8
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| static constexpr int FLAME_SLASH = 64; //火炎斬り | ||
| static constexpr int KACRACKLE_SLASH = 65;// マヒャド斬り | ||
| static constexpr int HATCHET_MAN = 66;// 魔人切り | ||
| static constexpr int UPWARD_SLICE = 67;// 切り上げとか | ||
| static constexpr int MULTISLASH = 68;// さみだれ斬り(さみだれ突きとは別行動) |
There was a problem hiding this comment.
These new IDs are now emitted by the vision pipeline, but the battle simulator does not implement them in BattleEmulator::callAttackFun (the switch falls to default, logs error!!!!!, and returns zero damage). In practice, once these recognitions fire, replay/search results diverge because the recognized action stream contains IDs that the simulator cannot execute.
Useful? React with 👍 / 👎.
| /public/vision-assets.js | ||
| /public/vision-assets.json |
There was a problem hiding this comment.
Ignoring both public/vision-assets.js and public/vision-assets.json after removing the tracked JSON leaves a fresh checkout with no asset pack files. Since index.html now loads vision-assets.js and the JS fallback fetches vision-assets.json, both paths fail without a separate build step, and initialization degrades to empty fallback templates, effectively disabling recognition in direct public/ runtime setups.
Useful? React with 👍 / 👎.
| static constexpr int POISON_ATTACK = 69; // 毒攻撃 | ||
| static constexpr int DECELERATLE = 70; //ボミオス | ||
| static constexpr int KASAP = 71; //ルカナン | ||
| static constexpr int SWEET_BREATH = 72; //ルカナン | ||
|
|
There was a problem hiding this comment.
This change shifts pre-existing action constants (starting here) to new numeric values instead of appending new IDs. Because the action protocol is numeric, previously captured/replayed action strings are reinterpreted as different moves under the new mapping, which breaks reproducibility and compatibility with existing logs/inputs.
Useful? React with 👍 / 👎.
| 64: {names: {ja: "火炎斬り", en: "FLAME_SLASH"}, ally: false, damage: true}, | ||
| 65: {names: {ja: "マヒャド斬り", en: "KACRACKLE_SLASH"}, ally: false, damage: true}, | ||
| 66: {names: {ja: "魔人切り", en: "HATCHET_MAN"}, ally: false, damage: true}, | ||
| 67: {names: {ja: "斬り上げた", en: "UPWARD_SLICE"}, ally: false, damage: true}, | ||
| 68: {names: {ja: "さみだれ斬り", en: "MULTISLASH"}, ally: false, damage: true} |
There was a problem hiding this comment.
buildBattleActionFormat routes IDs into enemy/ally buckets based on ACTIONS[actionId].ally, but these new slash IDs are emitted from the Gilyumei ally-side recognition rules and are currently marked ally: false. When they are detected, the generated battle format string places them on the enemy side, producing incorrect command payloads for downstream simulation.
Useful? React with 👍 / 👎.
1 participant
No description provided.