Resolve spacetime client binding merge conflicts

This commit is contained in:
2026-04-24 14:44:46 +08:00
parent 4f369617c7
commit f65177b147
26 changed files with 2172 additions and 1020 deletions

2
.idea/.name generated
View File

@@ -1 +1 @@
PreGameSelectionFlow.tsx
mod.rs

View File

@@ -83,3 +83,31 @@ cargo check -p spacetime-module
```
结果:通过。`spacetime-module` 仅保留仓库既有 glob re-export warning。
## 2026-04-24 追加:结果页删除与资产动作闭环
本次继续补齐除长尾补全外的结果页可见动作:
1. `delete_characters` / `delete_landmarks` 已由 Rust SpacetimeDB reducer 直接更新 `draft_profile_json``result_preview_json``publish_gate_json``checkpoints_json``custom_world_draft_card``custom_world_agent_operation``custom_world_agent_message`
2. `generate_characters` 增加 `roleType`,可扮演角色写入 `playableNpcs`,场景角色写入 `storyNpcs`,不再把可扮演角色落到场景角色列表。
3. `generate_role_assets` / `generate_scene_assets` 不再走占位动作Rust 会校验目标对象、切到 `visual_refining`、设置 `focus_card_id`,并记录 operation/message。
4. `sync_role_assets` / `sync_scene_assets` 已迁移 Node 的 profile 字段写回逻辑:角色写回 `imageSrc``generatedVisualAssetId``generatedAnimationSetId``animationMap`;场景写回 `imageSrc``generatedSceneAssetId``generatedScenePrompt``generatedSceneModel`,并同步 `sceneChapters.acts` 背景字段。
5. 前端结果页角色卡展示“生成资产”,场景卡展示“生成场景图”,均通过 `autosaveCoordinator.executeAgentActionAndWait` 调 Rust API 并用最新 session 重建预览。
本轮仍不迁移 `expand_long_tail`,保持后续单独设计。
## 2026-04-24 追加:创作 Tab 删除作品入口
用户在 `http://127.0.0.1:3000/` 的“创作”Tab 看不到删除作品入口,原因是 `RpgEntryHomeView``CreationLibraryCard` 只支持整卡打开详情,没有接收删除回调。已补齐:
1. `CreationLibraryCard` 右上角展示“删除”按钮,点击时阻止整卡打开详情。
2. `RpgEntryHomeView` 新增 `onDeleteLibraryEntry``deletingLibraryEntryId` props。
3. `PlatformEntryFlowShellImpl` 复用 `deleteRpgEntryWorldProfile`,删除后刷新我的作品列表与公开广场。
链路保持为:前端创作 Tab -> `deleteRpgEntryWorldProfile` -> Rust runtime API -> SpacetimeDB 软删除 profile / 移除 gallery 读模型。
## 2026-04-24 追加:创作 Hub 草稿删除入口修正
截图中的“创作”Tab 实际渲染的是 `CustomWorldCreationHub` / `CustomWorldWorkCard`,不是默认 `RpgEntryHomeView``CreationLibraryCard`。此前 Hub 只给 `status=published` 的 RPG 作品传入删除回调,导致草稿卡片没有“删除”按钮。
修正后:只要 RPG 创作条目存在 `profileId`,无论 `draft` 还是 `published`,都会在卡片底部动作区展示“删除”。删除继续复用 `PlatformEntryFlowShellImpl.handleDeletePublishedWork`,走 `deleteRpgEntryWorldProfile` -> Rust runtime API -> SpacetimeDB 软删除。

View File

@@ -52,3 +52,34 @@ cargo test -p api-server custom_world_foundation_draft --no-default-features
```
结果:`3 passed`
## 2026-04-24 进度链路补齐
本次继续补齐“点击生成世界草稿”后的异步执行方式,避免 HTTP 请求阻塞到全部 LLM 调用结束才返回:
1. `execute_custom_world_agent_action(draft_foundation)` 现在先创建 `draft_foundation` running operation并立即把 `operationId` 返回给前端。
2. API 后台任务继续执行 Node 同序多阶段生成;前端已有的 operation polling 可以持续读取阶段进度。
3. Rust 生成器按 Node 的 `onProgress` 节点写入:
- `12`:整理世界骨架。
- `16-30`:生成可扮演角色。
- `30-44`:生成场景角色。
- `44-56`:生成关键场景。
- `56-66`:建立场景连接。
- `66-76`:补全可扮演角色叙事基础。
- `76-84`:补全可扮演角色档案细节。
- `84-92`:补全场景角色叙事基础。
- `92-96`:补全场景角色档案细节。
- `97`:编译世界底稿。
- `98`:编译草稿卡。
4. SpacetimeDB 新增 `upsert_custom_world_agent_operation_progress`,只更新/创建 operation 进度,不插入聊天消息、不推进 turn专门承接生成中的阶段进度。
5. 最终落库仍复用 `execute_custom_world_agent_action(draft_foundation)`,但允许复用同一个 running operation 完成写入,避免中间断点和重复 operation。
补充验证:
```bash
cargo check -p api-server
cargo check -p spacetime-module
cargo test -p api-server custom_world_foundation_draft -- --nocapture
```
结果:后端检查通过;`custom_world_foundation_draft` 相关测试 `3 passed`

View File

@@ -26,26 +26,28 @@ Agent 结果页点击新增场景角色 / 新增场景
-> 刷新结果页 profile
```
说明:当前可扮演角色 tab 的“新增可扮演角色”会调用 `generate_characters`,后端现阶段会追加到 `storyNpcs`。因此严格意义上的“新增可扮演角色”仍未完整迁移,需要后续给 action 增加角色类型参数或新增 `generate_playable_characters`
说明:当前可扮演角色 tab 的“新增可扮演角色”会调用 `generate_characters` 并传入 `roleType=playable`Rust 会写入 `draftProfile.playableNpcs`;场景角色则写入 `draftProfile.storyNpcs`
## 已迁移 / 可见
1. 删除作品:已有 Rust + SpacetimeDB 软删除链路。
2. 新增场景角色:结果页可见,调用 Rust `generate_characters`
3. 新增场景 / 地点:结果页可见,调用 Rust `generate_landmarks`
4. Agent 结果页发布进入世界:已有 `publish_world` + publish gate 链路
5. 手动编辑结果页 profile目前仍通过 `sync_result_profile` 自动保存回 Agent session
2. 新增可扮演角色:结果页可见,调用 Rust `generate_characters(roleType=playable)` 并写入 `playableNpcs`
3. 新增场景角色:结果页可见,调用 Rust `generate_characters(roleType=story)` 并写入 `storyNpcs`
4. 新增场景 / 地点:结果页可见,调用 Rust `generate_landmarks` 并写入 `landmarks`
5. 批量删除场景角色:结果页可见,调用 Rust `delete_characters`,同步删除 profile 与 draft card
6. 批量删除场景:结果页可见,调用 Rust `delete_landmarks`,同步删除 profile、连接与 draft card。
7. 角色资产准备:结果页角色卡可见“生成资产”,调用 Rust `generate_role_assets`,进入 `visual_refining` 并聚焦角色。
8. 场景资产准备:结果页场景卡可见“生成场景图”,调用 Rust `generate_scene_assets`,进入 `visual_refining` 并聚焦场景。
9. 角色资产同步Rust `sync_role_assets` 会把 `portraitPath / generatedVisualAssetId / generatedAnimationSetId / animationMap` 写入 profile、draft card、asset coverage、preview、checkpoint、operation 与 message。
10. 场景资产同步Rust `sync_scene_assets` 会把 `imageSrc / generatedSceneAssetId / prompt / model` 写入 camp 或 landmark并同步 `sceneChapters.acts`、draft card、asset coverage、preview、checkpoint、operation 与 message。
11. Agent 结果页发布进入世界:已有 `publish_world` + publish gate 链路。
12. 手动编辑结果页 profile目前仍通过 `sync_result_profile` 自动保存回 Agent session。
## 尚未完整迁移的结果页编辑功能
1. 新增可扮演角色:前端有入口,但 Rust action 暂无角色类型区分,当前会落到 `storyNpcs`
2. 批量删除场景角色:前端只改本地 profile再靠 `sync_result_profile` 同步,不是独立 Rust action
3. 批量删除场景:前端只改本地 profile再靠 `sync_result_profile` 同步,不是独立 Rust action
4. 单个角色 / 场景的细粒度编辑:前端 modal 仍编辑本地 profile再靠 `sync_result_profile` 同步SpacetimeDB 虽有 `update_draft_card`,但结果页表单尚未按 card section action 化。
5. 角色资产生成:`generate_role_assets / sync_role_assets` Rust 侧仍是 placeholder 或外部链路未完全接入结果页。
6. 场景资产生成:`generate_scene_assets / sync_scene_assets` Rust 侧仍是 placeholder 或外部链路未完全接入结果页。
7. 长尾补全:`expand_long_tail` Rust 侧仍是 placeholder。
8. 回滚 checkpointRust 有 `revert_checkpoint`,但结果页没有清晰可见入口。
1. 单个角色 / 场景的细粒度编辑:前端 modal 仍编辑本地 profile再靠 `sync_result_profile` 同步SpacetimeDB 虽有 `update_draft_card`,但结果页表单尚未按 card section action 化
2. 长尾补全:`expand_long_tail` 本轮明确排除Rust 侧仍是 placeholder
3. 回滚 checkpointRust 有 `revert_checkpoint`,但结果页没有清晰可见入口
## 下一步建议

View File

@@ -0,0 +1,33 @@
# RPG 创作与资产提示词脚本抽离方案2026-04-24
## 背景
`server-rs/crates/api-server` 中 RPG 创作链路已经承接草稿生成、结果页补角色/补场景、场景图、角色图与角色动作生成。此前提示词散落在路由处理文件中,导致玩法规则、资产规则与结果页生成规则混杂,后续迭代容易出现落地漂移。
## 落地边界
本次只调整 Rust 后端 `api-server` 内的提示词组织,不兼容 `server-node`,也不改动前端展示文案。
## 模块拆分
1. `custom_world_rpg_draft_prompts.rs`
- 承载 RPG 玩法草稿生成相关提示词。
- 覆盖八锚点共创主提示词、状态识别提示词、模式规则、用户输入信号规则、上下文渲染。
- `custom_world_agent_turn.rs` 只保留流程编排、LLM 调用和结果规范化。
2. `custom_world_asset_prompts.rs`
- 承载生图、生动作相关提示词。
- 覆盖角色主图提示词、角色主图负面提示词、角色动作视频/序列帧提示词、动作兜底安全提示词。
- 场景图和封面图提示词也属于自定义世界资产提示词,统一迁入该模块。
3. `custom_world_result_prompts.rs`
- 承载结果页新增实体相关提示词。
- 覆盖新增可扮演角色、新增场景角色、新增场景的 LLM system/user prompt 构造。
- 路由层继续负责 fallback 与返回结构,提示词模块只负责生成可审计的 prompt 文本。
## 约束
- 提示词模块只做纯函数拼装,不访问网络、文件、数据库或 SpacetimeDB。
- 保留原中文提示词语义,不把中文改写成英文。
- 原有 fallback 行为不变LLM 不可用或解析失败时仍回退本地生成。
- 仅做局部迁移,避免整文件重写导致中文编码风险。
## 验证
- `cargo test -p api-server` 应能通过或至少完成编译阶段。
- 既有单元测试中关于角色图、动作结果 payload、场景图请求的断言应保持不变。

View File

@@ -43,7 +43,12 @@ use shared_contracts::assets::{
use spacetime_client::SpacetimeClientError;
use crate::{
api_response::json_success_body, http_error::AppError, request_context::RequestContext,
api_response::json_success_body,
custom_world_asset_prompts::{
build_character_animation_prompt, build_fallback_moderation_safe_animation_prompt,
},
http_error::AppError,
request_context::RequestContext,
state::AppState,
};
use tokio::time::sleep;
@@ -1713,310 +1718,11 @@ fn build_character_animation_job_payload(task: AiTaskSnapshot) -> CharacterAsset
}
}
fn build_character_animation_prompt(
strategy: &CharacterAnimationStrategy,
prompt_text: &str,
character_brief_text: Option<&str>,
action_template_id: Option<&str>,
animation: &str,
frame_count: u32,
fps: u32,
duration_seconds: u32,
loop_: bool,
use_chroma_key: bool,
) -> String {
match strategy {
CharacterAnimationStrategy::ImageToVideo => build_ark_character_animation_prompt(
animation,
prompt_text,
character_brief_text,
action_template_id,
loop_,
use_chroma_key,
),
CharacterAnimationStrategy::ImageSequence => {
build_image_sequence_prompt(animation, prompt_text, frame_count, use_chroma_key)
}
CharacterAnimationStrategy::MotionTransfer
| CharacterAnimationStrategy::ReferenceToVideo => build_npc_animation_prompt(
animation,
prompt_text,
character_brief_text,
action_template_id,
loop_,
use_chroma_key,
fps,
duration_seconds,
),
}
}
fn build_image_sequence_prompt(
animation: &str,
prompt_text: &str,
frame_count: u32,
use_chroma_key: bool,
) -> String {
[
format!(
"同一角色连续 {} 帧动作序列,动作主题是 {}",
frame_count, animation
),
"固定机位,单人,全身,侧身朝右,保持同一套服装、发型、武器和体型。".to_string(),
"帧间动作连续,姿态逐步推进,不要换人,不要跳变,不要多余物体。".to_string(),
if use_chroma_key {
"纯绿色背景,无地面装饰,方便后期抠像。".to_string()
} else {
"背景尽量纯净,避免复杂场景。".to_string()
},
prompt_text.trim().to_string(),
]
.into_iter()
.filter(|value| !value.trim().is_empty())
.collect::<Vec<_>>()
.join(" ")
}
fn build_npc_animation_prompt(
animation: &str,
prompt_text: &str,
character_brief_text: Option<&str>,
action_template_id: Option<&str>,
loop_: bool,
use_chroma_key: bool,
fps: u32,
duration_seconds: u32,
) -> String {
let character_brief = build_compact_animation_character_brief(character_brief_text);
let action_detail_text = sanitize_animation_prompt_text(prompt_text, 140);
let loop_rule = if loop_ {
"这是循环动作,直接进入动作循环中段,不要开场静止站桩,不要把主参考图原样作为第一帧。"
.to_string()
} else if animation == "die" {
"这是死亡终结动作,首帧参考主图角色形象即可,尾帧停在死亡结束姿态,不要回到主图形象。"
.to_string()
} else {
"这是非循环动作,首帧和尾帧都要回到参考主图角色形象,中段完成动作变化。".to_string()
};
if let Some(template) = action_template_id.and_then(|id| find_motion_template(id)) {
return [
format!(
"单人 NPC 全身动作视频,动作主题是 {}。角色固定为同一人,右向斜侧身,镜头稳定,轮廓清晰,武器不可丢失。",
template.animation
),
if use_chroma_key {
"背景为纯绿色绿幕,无其他人物和场景元素,方便后期抠像。".to_string()
} else {
"背景简洁纯净,无复杂场景。".to_string()
},
if character_brief.is_empty() {
String::new()
} else {
format!("角色设定:{}", character_brief)
},
format!("动作补充:{}", template.prompt_suffix),
if action_detail_text.is_empty() {
String::new()
} else {
format!("动作细节:{}", action_detail_text)
},
format!("目标帧率 {} fps时长约 {} 秒。", fps.clamp(1, 60), duration_seconds.clamp(1, 8)),
loop_rule,
]
.into_iter()
.filter(|value| !value.trim().is_empty())
.collect::<Vec<_>>()
.join(" ");
}
[
format!("单人 NPC 全身动作视频,动作主题是 {}", animation),
"角色固定为同一人,侧身朝右,镜头稳定,轮廓清晰,武器不可丢失。".to_string(),
"动作连贯,避免服装、发型、面部、武器随机漂移。".to_string(),
if use_chroma_key {
"背景为纯绿色绿幕,无其他人物和场景元素,方便后期抠像。".to_string()
} else {
"背景简洁纯净,无复杂场景。".to_string()
},
if character_brief.is_empty() {
String::new()
} else {
format!("角色设定:{}", character_brief)
},
if action_detail_text.is_empty() {
String::new()
} else {
action_detail_text
},
format!(
"目标帧率 {} fps时长约 {} 秒。",
fps.clamp(1, 60),
duration_seconds.clamp(1, 8)
),
loop_rule,
]
.into_iter()
.filter(|value| !value.trim().is_empty())
.collect::<Vec<_>>()
.join(" ")
}
fn build_ark_character_animation_prompt(
animation: &str,
prompt_text: &str,
character_brief_text: Option<&str>,
action_template_id: Option<&str>,
loop_: bool,
use_chroma_key: bool,
) -> String {
let normalized_animation_name = animation.trim().replace(char::is_whitespace, "_");
let normalized_animation_name = if normalized_animation_name.is_empty() {
"idle".to_string()
} else {
normalized_animation_name
};
let character_brief = build_compact_animation_character_brief(character_brief_text);
let action_detail_text = sanitize_animation_prompt_text(prompt_text, 140);
let frame_rule = if loop_ {
"首帧严格使用图片1尾帧严格使用图片2循环动作必须自然闭环不要静止开场。".to_string()
} else {
"首帧严格使用图片1尾帧严格使用图片2中段完成完整动作变化收束干净。".to_string()
};
if let Some(template) = action_template_id.and_then(find_motion_template) {
return [
format!(
"单人 NPC 全身动作视频,动作英文名是 {}。角色固定为图片1和图片2中的同一人侧身朝右镜头稳定轮廓清晰武器不可丢失。",
normalized_animation_name
),
"动作连贯,避免服装、发型、面部、武器随机漂移,不要多角色,不要镜头切换。".to_string(),
if use_chroma_key {
"背景为纯绿色绿幕,无其他人物和场景元素,方便后期抠像。".to_string()
} else {
"背景简洁纯净,无复杂场景。".to_string()
},
if character_brief.is_empty() {
String::new()
} else {
format!("角色设定:{}", character_brief)
},
format!("动作补充:{}", template.prompt_suffix),
if action_detail_text.is_empty() {
String::new()
} else {
format!("动作细节:{}", action_detail_text)
},
frame_rule,
]
.into_iter()
.filter(|value| !value.trim().is_empty())
.collect::<Vec<_>>()
.join(" ");
}
[
format!(
"单人 NPC 全身动作视频,动作英文名是 {}",
normalized_animation_name
),
"角色固定为图片1和图片2中的同一人侧身朝右镜头稳定轮廓清晰武器不可丢失。"
.to_string(),
"动作连贯,避免服装、发型、面部、武器随机漂移,不要多角色,不要镜头切换。".to_string(),
if use_chroma_key {
"背景为纯绿色绿幕,无其他人物和场景元素,方便后期抠像。".to_string()
} else {
"背景简洁纯净,无复杂场景。".to_string()
},
if character_brief.is_empty() {
String::new()
} else {
format!("角色设定:{}", character_brief)
},
if action_detail_text.is_empty() {
String::new()
} else {
format!("动作细节:{}", action_detail_text)
},
frame_rule,
]
.into_iter()
.filter(|value| !value.trim().is_empty())
.collect::<Vec<_>>()
.join(" ")
}
fn build_fallback_moderation_safe_animation_prompt(
animation: &str,
loop_: bool,
use_chroma_key: bool,
) -> String {
[
format!("单人全身角色动作视频,动作主题是 {}", animation),
"角色固定为同一人,右向斜侧身,镜头稳定,轮廓清楚。".to_string(),
if loop_ {
"循环动作直接进入稳定循环,不要静止开场,不要定格首帧。".to_string()
} else {
"非循环动作首尾回到角色标准站姿,中段完成动作变化。".to_string()
},
if use_chroma_key {
"背景为纯绿色绿幕,无其他人物和场景元素。".to_string()
} else {
"背景简洁纯净。".to_string()
},
]
.join(" ")
}
fn sanitize_animation_prompt_text(value: &str, max_length: usize) -> String {
value
.replace(char::is_whitespace, " ")
.replace("血浆", "")
.replace("喷血", "")
.replace("鲜血", "")
.replace("断肢", "")
.replace("斩首", "")
.replace("裸体", "")
.replace("裸露", "")
.replace("色情", "")
.replace("性交", "")
.replace("死亡", "倒地结束")
.replace("死去", "倒地结束")
.replace("击杀", "倒地结束")
.replace("受击", "失衡")
.replace("受伤", "失衡")
.replace("砍杀", "挥击")
.replace("斩击", "挥击")
.split_whitespace()
.collect::<Vec<_>>()
.join(" ")
.chars()
.take(max_length)
.collect::<String>()
.trim()
.to_string()
}
fn build_compact_animation_character_brief(value: Option<&str>) -> String {
let normalized = sanitize_animation_prompt_text(value.unwrap_or_default(), 160);
if normalized.is_empty() {
return String::new();
}
normalized
.split(['/', '|', '\n', '', ',', '。', '', ';'])
.map(str::trim)
.filter(|item| !item.is_empty())
.take(4)
.collect::<Vec<_>>()
.join("")
}
fn find_motion_template(id: &str) -> Option<&'static MotionTemplate> {
pub(crate) fn find_motion_template(id: &str) -> Option<&'static MotionTemplate> {
BUILT_IN_MOTION_TEMPLATES
.iter()
.find(|template| template.id == id.trim())
}
fn resolve_character_animation_model(payload: &CharacterAnimationGenerateRequest) -> String {
let candidate = match payload.strategy {
CharacterAnimationStrategy::ImageSequence => payload.image_sequence_model.as_str(),
@@ -3486,12 +3192,12 @@ fn character_animation_error_response(
error.into_response_with_context(Some(request_context))
}
struct MotionTemplate {
id: &'static str,
label: &'static str,
animation: &'static str,
prompt_suffix: &'static str,
notes: &'static str,
pub(crate) struct MotionTemplate {
pub(crate) id: &'static str,
pub(crate) label: &'static str,
pub(crate) animation: &'static str,
pub(crate) prompt_suffix: &'static str,
pub(crate) notes: &'static str,
}
impl MotionTemplate {
@@ -3677,6 +3383,11 @@ mod tests {
}
#[test]
pub(crate) fn find_motion_template(id: &str) -> Option<&'static MotionTemplate> {
BUILT_IN_MOTION_TEMPLATES
.iter()
.find(|template| template.id == id.trim())
}
fn resolve_character_animation_model_uses_strategy_specific_field() {
let payload = CharacterAnimationGenerateRequest {
character_id: "hero".to_string(),

View File

@@ -32,7 +32,12 @@ use shared_contracts::assets::{
use spacetime_client::SpacetimeClientError;
use crate::{
api_response::json_success_body, http_error::AppError, request_context::RequestContext,
api_response::json_success_body,
custom_world_asset_prompts::{
build_character_visual_negative_prompt, build_character_visual_prompt,
},
http_error::AppError,
request_context::RequestContext,
state::AppState,
};
use tokio::time::sleep;
@@ -671,58 +676,8 @@ fn build_character_visual_job_payload(task: AiTaskSnapshot) -> CharacterAssetJob
}
}
fn build_character_visual_prompt(prompt_text: &str, character_brief_text: Option<&str>) -> String {
let merged = [character_brief_text.unwrap_or_default(), prompt_text]
.into_iter()
.map(str::trim)
.filter(|value| !value.is_empty())
.collect::<Vec<_>>()
.join("\n");
format!(
"{}\n单人全身右向斜侧身3 到 4 头身,像素动作角色,纯绿色背景,服装完整,轮廓清晰,不要复杂背景。",
if merged.is_empty() {
"自定义世界角色,服装完整,姿态自然。"
} else {
merged.as_str()
}
)
}
fn build_character_visual_negative_prompt() -> String {
[
"正面视角",
"左朝向",
"完全 90 度纯右视图",
"镜头透视",
"半身像",
"脚被裁切",
"头顶被裁切",
"多角色",
"复杂背景",
"建筑场景",
"漂浮物",
"烟雾环境",
"武器消失",
"武器换手",
"额外手臂",
"额外腿",
"服装变化",
"脸部变化",
"模糊",
"运动模糊",
"文字",
"水印",
"UI 元素",
"软萌 Q版大头贴",
"儿童绘本风",
"厚涂插画感",
"低对比柔边",
]
.join("")
}
fn require_dashscope_settings(state: &AppState) -> Result<DashScopeSettings, AppError> {
// Stage 2 的真实图片生成统一走 DashScope这里先把配置缺失拦在业务入口前。
let base_url = state.config.dashscope_base_url.trim().trim_end_matches('/');
if base_url.is_empty() {
return Err(
@@ -752,7 +707,6 @@ fn require_dashscope_settings(state: &AppState) -> Result<DashScopeSettings, App
request_timeout_ms: state.config.dashscope_image_request_timeout_ms.max(1),
})
}
fn build_dashscope_http_client(settings: &DashScopeSettings) -> Result<reqwest::Client, AppError> {
reqwest::Client::builder()
.timeout(Duration::from_millis(settings.request_timeout_ms))

View File

@@ -29,10 +29,10 @@ use shared_kernel::build_prefixed_uuid_id;
use spacetime_client::{
CustomWorldAgentActionExecuteRecordInput, CustomWorldAgentCheckpointRecord,
CustomWorldAgentMessageRecord, CustomWorldAgentMessageSubmitRecordInput,
CustomWorldAgentOperationRecord, CustomWorldAgentSessionCreateRecordInput,
CustomWorldAgentSessionRecord, CustomWorldDraftCardDetailRecord,
CustomWorldDraftCardDetailSectionRecord, CustomWorldDraftCardRecord,
CustomWorldGalleryEntryRecord, CustomWorldLibraryEntryRecord,
CustomWorldAgentOperationProgressRecordInput, CustomWorldAgentOperationRecord,
CustomWorldAgentSessionCreateRecordInput, CustomWorldAgentSessionRecord,
CustomWorldDraftCardDetailRecord, CustomWorldDraftCardDetailSectionRecord,
CustomWorldDraftCardRecord, CustomWorldGalleryEntryRecord, CustomWorldLibraryEntryRecord,
CustomWorldProfileUpsertRecordInput, CustomWorldPublishGateRecord,
CustomWorldResultPreviewBlockerRecord, CustomWorldSupportedActionRecord,
CustomWorldWorkSummaryRecord, SpacetimeClientError,
@@ -925,39 +925,44 @@ pub async fn execute_custom_world_agent_action(
})),
));
}
let draft_result = generate_custom_world_foundation_draft(llm_client, &session)
let operation_id = build_prefixed_uuid_id("operation-");
let operation = state
.spacetime_client()
.upsert_custom_world_agent_operation_progress(
CustomWorldAgentOperationProgressRecordInput {
session_id: session_id.clone(),
owner_user_id: owner_user_id.clone(),
operation_id: operation_id.clone(),
operation_type: "draft_foundation".to_string(),
operation_status: "running".to_string(),
phase_label: "整理世界骨架".to_string(),
phase_detail: "正在校验已确认锚点,并准备第一版世界框架生成链路。"
.to_string(),
operation_progress: 12,
error_message: None,
updated_at_micros: submitted_at_micros,
},
)
.await
.map_err(|message| {
.map_err(|error| {
custom_world_error_response(
&request_context,
AppError::from_status(StatusCode::BAD_GATEWAY).with_details(json!({
"provider": "custom-world-agent",
"message": message,
})),
map_custom_world_client_error(error),
)
})?;
build_draft_foundation_action_payload_json(&payload, &draft_result.draft_profile_json)
.map_err(|error| {
let (status, message) = match error {
DraftFoundationPayloadError::SerializePayload(message) => {
(StatusCode::BAD_REQUEST, message)
}
DraftFoundationPayloadError::InvalidPayloadShape => (
StatusCode::BAD_REQUEST,
"action payload 必须是 object".to_string(),
),
DraftFoundationPayloadError::InvalidGeneratedDraft(message) => {
(StatusCode::BAD_GATEWAY, message)
}
};
custom_world_error_response(
&request_context,
AppError::from_status(status).with_details(json!({
"provider": "custom-world-agent",
"message": message,
})),
)
})?
spawn_custom_world_draft_foundation_job(
state.clone(),
session,
owner_user_id,
operation_id,
payload,
);
return Ok(json_success_body(
Some(&request_context),
json!({
"operation": map_custom_world_agent_operation_response(operation),
}),
));
} else {
let generation_result =
generate_custom_world_agent_entities(llm_client, &session, &payload)
@@ -1021,6 +1026,177 @@ pub async fn execute_custom_world_agent_action(
))
}
fn spawn_custom_world_draft_foundation_job(
state: AppState,
session: CustomWorldAgentSessionRecord,
owner_user_id: String,
operation_id: String,
payload: ExecuteCustomWorldAgentActionRequest,
) {
tokio::spawn(async move {
let Some(llm_client) = state.llm_client().cloned() else {
let _ = upsert_custom_world_draft_foundation_progress(
&state,
&session.session_id,
&owner_user_id,
&operation_id,
"failed",
"底稿生成失败",
"服务端尚未配置可用的 LLM API Key",
100,
Some("服务端尚未配置可用的 LLM API Key".to_string()),
)
.await;
return;
};
let progress_state = state.clone();
let progress_session_id = session.session_id.clone();
let progress_owner_user_id = owner_user_id.clone();
let progress_operation_id = operation_id.clone();
let draft_result =
generate_custom_world_foundation_draft(&llm_client, &session, move |progress| {
let progress_state = progress_state.clone();
let session_id = progress_session_id.clone();
let owner_user_id = progress_owner_user_id.clone();
let operation_id = progress_operation_id.clone();
tokio::spawn(async move {
let _ = upsert_custom_world_draft_foundation_progress(
&progress_state,
&session_id,
&owner_user_id,
&operation_id,
"running",
progress.phase_label.as_str(),
progress.phase_detail.as_str(),
progress.progress,
None,
)
.await;
});
})
.await;
let draft_result = match draft_result {
Ok(result) => result,
Err(message) => {
let _ = upsert_custom_world_draft_foundation_progress(
&state,
&session.session_id,
&owner_user_id,
&operation_id,
"failed",
"底稿生成失败",
message.clone().as_str(),
100,
Some(message),
)
.await;
return;
}
};
let _ = upsert_custom_world_draft_foundation_progress(
&state,
&session.session_id,
&owner_user_id,
&operation_id,
"running",
"编译草稿卡",
"正在把世界底稿整理成可浏览的卡片摘要和详情结构。",
98,
None,
)
.await;
let payload_json = match build_draft_foundation_action_payload_json(
&payload,
&draft_result.draft_profile_json,
) {
Ok(value) => value,
Err(error) => {
let message = match error {
DraftFoundationPayloadError::SerializePayload(message) => message,
DraftFoundationPayloadError::InvalidPayloadShape => {
"action payload 必须是 object".to_string()
}
DraftFoundationPayloadError::InvalidGeneratedDraft(message) => message,
};
let _ = upsert_custom_world_draft_foundation_progress(
&state,
&session.session_id,
&owner_user_id,
&operation_id,
"failed",
"底稿写入失败",
message.clone().as_str(),
100,
Some(message),
)
.await;
return;
}
};
if let Err(error) = state
.spacetime_client()
.execute_custom_world_agent_action(CustomWorldAgentActionExecuteRecordInput {
session_id: session.session_id.clone(),
owner_user_id: owner_user_id.clone(),
operation_id: operation_id.clone(),
action: "draft_foundation".to_string(),
payload_json: Some(payload_json),
submitted_at_micros: current_utc_micros(),
})
.await
{
let message = error.to_string();
let _ = upsert_custom_world_draft_foundation_progress(
&state,
&session.session_id,
&owner_user_id,
&operation_id,
"failed",
"底稿写入失败",
message.clone().as_str(),
100,
Some(message),
)
.await;
}
});
}
async fn upsert_custom_world_draft_foundation_progress(
state: &AppState,
session_id: &str,
owner_user_id: &str,
operation_id: &str,
status: &str,
phase_label: &str,
phase_detail: &str,
progress: u32,
error_message: Option<String>,
) -> Result<CustomWorldAgentOperationRecord, SpacetimeClientError> {
state
.spacetime_client()
.upsert_custom_world_agent_operation_progress(
CustomWorldAgentOperationProgressRecordInput {
session_id: session_id.to_string(),
owner_user_id: owner_user_id.to_string(),
operation_id: operation_id.to_string(),
operation_type: "draft_foundation".to_string(),
operation_status: status.to_string(),
phase_label: phase_label.to_string(),
phase_detail: phase_detail.to_string(),
operation_progress: progress.min(100),
error_message,
updated_at_micros: current_utc_micros(),
},
)
.await
}
fn map_custom_world_library_entry_response(
entry: CustomWorldLibraryEntryRecord,
) -> CustomWorldLibraryEntryResponse {

View File

@@ -5,6 +5,14 @@ use module_custom_world::{
use platform_llm::{LlmClient, LlmMessage, LlmStreamDelta, LlmTextRequest};
use serde::{Deserialize, Serialize};
use serde_json::{Value as JsonValue, json};
use crate::custom_world_rpg_draft_prompts::{
BASE_SYSTEM_PROMPT, GLOBAL_HARD_RULES, OUTPUT_CONTRACT_REMINDER, QUICK_FILL_EXTRA_RULES,
STATE_INFERENCE_OUTPUT_CONTRACT, STATE_INFERENCE_SYSTEM_PROMPT,
extract_reply_text_from_partial_json, mode_rules, parse_conversation_mode, parse_drift_risk,
parse_json_response_text, parse_user_input_signal, render_chat_history_context,
render_current_anchor_context, render_dynamic_state_context, user_signal_rules,
};
use spacetime_client::{
CustomWorldAgentMessageFinalizeRecordInput, CustomWorldAgentMessageRecord,
CustomWorldAgentSessionRecord,
@@ -42,7 +50,7 @@ pub(crate) struct CustomWorldAgentTurnResult {
}
#[derive(Clone, Copy, Debug, PartialEq, Eq)]
enum PromptUserInputSignal {
pub(crate) enum PromptUserInputSignal {
Rich,
Normal,
Sparse,
@@ -51,14 +59,14 @@ enum PromptUserInputSignal {
}
#[derive(Clone, Copy, Debug, PartialEq, Eq)]
enum PromptDriftRisk {
pub(crate) enum PromptDriftRisk {
Low,
Medium,
High,
}
#[derive(Clone, Copy, Debug, PartialEq, Eq)]
enum PromptConversationMode {
pub(crate) enum PromptConversationMode {
Bootstrap,
Expand,
Compress,
@@ -69,18 +77,18 @@ enum PromptConversationMode {
#[derive(Clone, Debug)]
#[allow(dead_code)]
struct PromptDynamicState {
pub(crate) struct PromptDynamicState {
current_turn: u32,
progress_percent: u32,
user_input_signal: PromptUserInputSignal,
drift_risk: PromptDriftRisk,
pub(crate) user_input_signal: PromptUserInputSignal,
pub(crate) drift_risk: PromptDriftRisk,
quick_fill_requested: bool,
conversation_mode: PromptConversationMode,
judgement_summary: String,
pub(crate) conversation_mode: PromptConversationMode,
pub(crate) judgement_summary: String,
}
#[derive(Clone, Debug, Default)]
struct PromptDynamicStateInference {
pub(crate) struct PromptDynamicStateInference {
user_input_signal: Option<PromptUserInputSignal>,
drift_risk: Option<PromptDriftRisk>,
conversation_mode: Option<PromptConversationMode>,
@@ -177,7 +185,7 @@ struct IconicElementValue {
#[derive(Clone, Debug, Default, Serialize, Deserialize)]
#[serde(rename_all = "camelCase")]
struct EightAnchorContent {
pub(crate) struct EightAnchorContent {
#[serde(default)]
world_promise: Option<WorldPromiseValue>,
#[serde(default)]
@@ -271,229 +279,6 @@ impl std::fmt::Display for CustomWorldTurnError {
impl std::error::Error for CustomWorldTurnError {}
const BASE_SYSTEM_PROMPT: &str = r#"你是一个负责共创游戏世界设定的专业策划。
你正在和用户一起共创一个游戏世界。每一轮你都必须读取:
1. 当前完整设定结构
2. 用户聊天记录
然后输出:
1. 一版新的完整设定结构
2. 当前 progress 百分比
3. 一段直接回复用户的话
你必须把“新的完整设定结构”视为下一轮的唯一有效版本。
你的输出会直接覆盖上一版设定结构。
你不是在做局部 patch。
你不是在做解释报告。
你不是在给开发者写分析。
你是在同时完成:
1. 世界设定更新
2. 当前推进程度判断
3. 对用户的共创回复"#;
const GLOBAL_HARD_RULES: &str = r#"全局硬约束:
1. 必须输出完整的设定结构,而不是只输出变化部分。
2. 新的设定结构会直接覆盖旧内容,因此不得随意丢失仍然成立的重要信息。
3. 如果用户明确修正旧设定,必须在新的设定结构中直接体现修正结果。
4. 如果用户输入信息不足,可以保留上一版中仍然成立的内容。
5. progressPercent 最低为 0不允许为负数。
6. replyText 会直接发送给用户,因此要自然、直接、可继续聊天。
7. 不要输出额外解释,不要输出 markdown 代码块,不要输出开发备注。
8. replyText 不要写成长篇策划文,不要展开大段世界观百科。
9. replyText 默认只推进当前最关键的一步,不要同时抛出很多话题。
10. replyText 不要提及“八锚点”“锚点”“结构字段”“框架字段”等内部概念词。
11. 你输出的 JSON 必须可以被直接解析。
12. 输出字段顺序必须固定为replyText、progressPercent、nextAnchorContent。"#;
const QUICK_FILL_EXTRA_RULES: &str = r#"用户刚刚主动要求你自动补全剩余设定。
这表示用户接受你基于当前方向自动补完剩余设定。
本轮要求:
1. 不要再继续提问
2. 直接输出一版尽量完整的设定结构
3. progressPercent 直接输出为 100
4. replyText 要告诉用户现在可以进入“生成游戏设定草稿”"#;
const STATE_INFERENCE_SYSTEM_PROMPT: &str = r#"你是正式生成世界设定前的一步“创作状态识别器”。
你的职责不是直接生成新设定,而是先判断:下一轮正式生成应该用什么推进策略,尤其要判断 replyText 应该更偏确认、吸收、收束、纠偏,还是启发式提问。
你必须综合以下信息判断:
1. 当前轮次 currentTurn
2. 当前完成度 progressPercent
3. 用户是否要求自动补全 quickFillRequested
4. 当前完整设定结构
5. 最近聊天记录,尤其是最近 1 到 3 轮用户消息
你需要输出 4 个字段:
1. userInputSignal只能是 rich / normal / sparse / correction / delegate
2. driftRisk只能是 low / medium / high
3. conversationMode只能是 bootstrap / expand / compress / repair_direction / force_complete / closing
4. judgementSummary1 到 2 句中文,概括你为什么这样判断,以及正式生成时最该注意什么
请按下面的语义判断。
一、userInputSignal 定义
1. rich
- 用户这一轮给了多条可直接落地的有效信息
- 这些信息可能同时覆盖世界方向、玩家处境、开局事件、冲突、关系、标志元素中的多个
- 正式生成时应优先高密度吸收,不要只更新一个点
2. normal
- 用户在顺着当前方向做正常补充
- 信息量中等,有明确新增内容,但没有明显推翻旧方向,也没有把决定权交给系统
- 正式生成时应稳定推进并自然接住用户内容
3. sparse
- 用户输入很短、很虚、很笼统,或几乎没有新增有效事实
- 例如只有一个题材词、一个气质词、一句很概括的话、一个很短的倾向表达
- 这种情况下,正式生成阶段的 replyText 应优先采用启发式提问
- 启发式提问的要求是:只问一个最容易回答、最能推动落地设计的问题
4. correction
- 用户这轮核心动作是在修正、替换、推翻、重定向旧设定
- 即使文字不长,只要主意图是“之前那个不对,现在改成这个”,也应优先判为 correction
- correction 的优先级高于 rich 和 normal
5. delegate
- 用户把部分决定权交给系统
- 例如“你来定”“你帮我补”“按你觉得合理的来”“先给我一个默认方案”
- delegate 关注的是授权关系,不只是信息多寡
二、driftRisk 定义
1. low
- 当前轮输入与已有方向基本一致
- 没有明显改口或冲突
2. medium
- 当前轮带来一定方向变化或扩张
- 还没有明显推翻旧方向,但如果处理不好,容易让设定开始发散
3. high
- 用户明确纠偏、改口、替换方向,或最近多轮反复修正
- 这时最重要的是防止旧方向重新回流到正式生成结果里
三、conversationMode 选择原则
1. bootstrap
- 适用于前期、信息少、核心方向未稳定
- replyText 更适合低压力确认和单点启发
2. expand
- 适用于方向已成形,正在顺着现有路线继续补充
- replyText 更适合总结已接住的内容并往前推一步
3. compress
- 适用于中后段,已有骨架,需要开始收束
- replyText 更适合聚焦最关键缺口,而不是继续开支线
4. repair_direction
- 适用于用户正在纠偏
- replyText 更适合先承认修正,再沿修正后的方向继续推进
5. force_complete
- 适用于用户明确要求自动补全
- replyText 不再提问,而应给出完成感和下一步引导
6. closing
- 适用于接近完成但并非强制一键补全
- replyText 更像确认与收束,而不是前期式探索
四、优先级规则
1. 如果 quickFillRequested 为 trueconversationMode 必须优先判为 force_complete
2. 如果用户核心意图是修正旧方向userInputSignal 优先判为 correctionconversationMode 通常优先考虑 repair_direction
3. 如果用户核心意图是授权系统替他补完userInputSignal 优先判为 delegate
4. 只有在没有明显纠偏、也没有明确自动补全要求时,才主要依据 currentTurn、progressPercent 和信息密度,在 bootstrap / expand / compress / closing 之间选择
五、关于 replyText 风格的专门判断要求
1. 如果用户输入较少、较虚或不够落地,正式生成阶段的 replyText 应采用启发式提问
2. 启发式提问一次最多只能提 1 个问题,不能连问两个或更多
3. 启发式提问必须问“最能推动当前设计落地”的那个问题,而不是泛泛而谈
4. 如果用户输入已经足够 rich就不要再机械提问优先吸收和推进
5. 如果用户在 correction 或 delegate 状态下replyText 是否提问要服从更高目标:纠偏生效或代为补全,不要机械套 sparse 的问法
六、关于 replyText 用语的硬约束
1. replyText 禁止提及内部结构名、锚点名、字段名、schema 名、框架词
2. 禁止出现这类内部表达:世界承诺、玩家幻想、主题边界、玩家入口、核心冲突、关键关系、隐藏线、标志元素、字段、结构、模块、八锚点
3. replyText 只能用通俗、直接、面向创作沟通的语言回应用户
4. replyText 应该围绕用户正在讨论的具体内容来落地,比如身份、开场处境、冲突、人物关系、地点、规则、气质,而不是抽象谈结构
5. judgementSummary 可以简洁提到“这轮更适合启发式提问”或“这轮应优先吸收修正”,但也不要堆内部术语
七、关于 judgementSummary 的写法
1. 必须简洁,不要写成长篇分析
2. 必须直接服务于下一轮正式生成
3. 最好同时包含两层信息:
- 为什么这么判断
- 正式生成时最该优先做什么,或最该避免什么
八、硬性约束
1. 只能输出 JSON不能输出解释、代码块或额外说明
2. 不能发明上下文里不存在的设定事实
3. 你的任务是“判断生成策略”,不是“代替正式生成直接写新设定”
4. 即使信息不完全,也必须在给定枚举里选出最合理的一组状态
5. judgementSummary 必须是中文
6. 输出值必须严格落在给定枚举中"#;
const STATE_INFERENCE_OUTPUT_CONTRACT: &str = r#"请严格按以下 JSON 结构输出,不要输出其他文字:
{
"userInputSignal": "normal",
"driftRisk": "low",
"conversationMode": "expand",
"judgementSummary": ""
}"#;
const OUTPUT_CONTRACT_REMINDER: &str = r#"请严格按以下 JSON 结构输出,不要输出其他文字:
{
"replyText": "",
"progressPercent": 0,
"nextAnchorContent": {
"worldPromise": {
"hook": "",
"differentiator": "",
"desiredExperience": ""
},
"playerFantasy": {
"playerRole": "",
"corePursuit": "",
"fearOfLoss": ""
},
"themeBoundary": {
"toneKeywords": [],
"aestheticDirectives": [],
"forbiddenDirectives": []
},
"playerEntryPoint": {
"openingIdentity": "",
"openingProblem": "",
"entryMotivation": ""
},
"coreConflict": {
"surfaceConflicts": [],
"hiddenCrisis": "",
"firstTouchedConflict": ""
},
"keyRelationships": [
{
"pairs": "",
"relationshipType": "",
"secretOrCost": ""
}
],
"hiddenLines": {
"hiddenTruths": [],
"misdirectionHints": [],
"revealPacing": ""
},
"iconicElements": {
"iconicMotifs": [],
"institutionsOrArtifacts": [],
"hardRules": []
}
}
}"#;
pub(crate) async fn run_custom_world_agent_turn<F>(
request: CustomWorldAgentTurnRequest<'_>,
on_reply_update: F,
@@ -1679,293 +1464,6 @@ fn summarize_dynamic_state(
)
}
fn render_dynamic_state_context(dynamic_state: &PromptDynamicState) -> String {
format!(
"上一轮预判得到的创作状态如下。\n正式生成时必须把它作为本轮策略输入直接执行,不要重新另起一套判断。\n\n创作状态:\n- userInputSignal: {}\n- driftRisk: {}\n- conversationMode: {}\n- judgementSummary: {}",
dynamic_state.user_input_signal.as_str(),
dynamic_state.drift_risk.as_str(),
dynamic_state.conversation_mode.as_str(),
dynamic_state.judgement_summary
)
}
fn render_current_anchor_context(anchor_content: &EightAnchorContent) -> String {
format!(
"当前完整设定结构如下。\n你必须把它视为上一版有效世界底子。\n\n如果用户没有否定其中某部分内容,且该部分仍然成立,可以继续保留。\n如果用户明确修正了某部分内容,新的完整设定结构必须体现修正后的版本。\n\n当前完整设定结构:\n{}",
serde_json::to_string_pretty(anchor_content)
.unwrap_or_else(|_| empty_agent_anchor_content_json())
)
}
fn render_chat_history_context(chat_history: &[JsonValue]) -> String {
format!(
"以下是用户聊天记录。\n请重点理解最近几轮里用户新增、修正、强调的设定信息。\n不要把早期已经被用户否定的内容继续当成最终结论。\n\n用户聊天记录:\n{}",
serde_json::to_string_pretty(chat_history).unwrap_or_else(|_| "[]".to_string())
)
}
fn parse_json_response_text(text: &str) -> Result<JsonValue, serde_json::Error> {
let trimmed = text.trim();
if let Some(start) = trimmed.find('{')
&& let Some(end) = trimmed.rfind('}')
&& end > start
{
return serde_json::from_str::<JsonValue>(&trimmed[start..=end]);
}
serde_json::from_str::<JsonValue>(trimmed)
}
fn extract_reply_text_from_partial_json(text: &str) -> Option<String> {
let key_index = text.find("\"replyText\"")?;
let colon_index = text[key_index..].find(':')? + key_index;
let mut cursor = colon_index + 1;
while cursor < text.len() && text.as_bytes()[cursor].is_ascii_whitespace() {
cursor += 1;
}
if text.as_bytes().get(cursor).copied() != Some(b'"') {
return None;
}
cursor += 1;
let mut decoded = String::new();
let remainder = text.get(cursor..)?;
let mut characters = remainder.chars().peekable();
while let Some(current) = characters.next() {
if current == '"' {
return Some(decoded);
}
if current == '\\' {
let escaped = characters.next()?;
match escaped {
'"' => decoded.push('"'),
'\\' => decoded.push('\\'),
'/' => decoded.push('/'),
'b' => decoded.push('\u{0008}'),
'f' => decoded.push('\u{000C}'),
'n' => decoded.push('\n'),
'r' => decoded.push('\r'),
't' => decoded.push('\t'),
'u' => {
let mut hex = String::new();
for _ in 0..4 {
hex.push(characters.next()?);
}
if let Ok(code) = u16::from_str_radix(hex.as_str(), 16)
&& let Some(character) = char::from_u32(code as u32)
{
decoded.push(character);
}
}
other => decoded.push(other),
}
continue;
}
decoded.push(current);
}
Some(decoded)
}
fn parse_user_input_signal(value: Option<&JsonValue>) -> Option<PromptUserInputSignal> {
match value.and_then(JsonValue::as_str)? {
"rich" => Some(PromptUserInputSignal::Rich),
"normal" => Some(PromptUserInputSignal::Normal),
"sparse" => Some(PromptUserInputSignal::Sparse),
"correction" => Some(PromptUserInputSignal::Correction),
"delegate" => Some(PromptUserInputSignal::Delegate),
_ => None,
}
}
fn parse_drift_risk(value: Option<&JsonValue>) -> Option<PromptDriftRisk> {
match value.and_then(JsonValue::as_str)? {
"low" => Some(PromptDriftRisk::Low),
"medium" => Some(PromptDriftRisk::Medium),
"high" => Some(PromptDriftRisk::High),
_ => None,
}
}
fn parse_conversation_mode(value: Option<&JsonValue>) -> Option<PromptConversationMode> {
match value.and_then(JsonValue::as_str)? {
"bootstrap" => Some(PromptConversationMode::Bootstrap),
"expand" => Some(PromptConversationMode::Expand),
"compress" => Some(PromptConversationMode::Compress),
"repair_direction" => Some(PromptConversationMode::RepairDirection),
"force_complete" => Some(PromptConversationMode::ForceComplete),
"closing" => Some(PromptConversationMode::Closing),
_ => None,
}
}
fn mode_rules(mode: PromptConversationMode) -> &'static str {
match mode {
PromptConversationMode::Bootstrap => {
r#"当前模式bootstrap
目标:
1. 先把世界的基本方向抓住
2. 不要一次塞太多新设定
3. 回复要降低用户开口压力
本轮行为要求:
1. 优先从用户输入里抓世界方向、玩家视角、主题边界的线索
2. 如果用户信息很少,不要强行把整套结构一次补满
3. replyText 要像共创搭档,而不是像审问
4. 默认只推进一个最关键的问题方向
5. 如果用户刚开口,优先给“被理解感”,再轻轻推进下一步
6. 可以用一句很短的话先确认你抓到的核心方向,再提一个最好回答的问题
7. 不要把问题问得像表单采集,不要一口气追问多个维度
用户体验要求:
1. 让用户觉得“现在很容易继续往下说”
2. 不要制造被考试、被拷问、被策划问卷追着跑的感觉
3. replyText 最好短、稳、可接话
4. 如果用户信息很少,也不要显得冷淡或机械"#
}
PromptConversationMode::Expand => {
r#"当前模式expand
目标:
1. 在保持现有方向的前提下,把设定结构逐步补全
2. 尽量让一轮输入覆盖多个关键维度
本轮行为要求:
1. 继续保留上一版里仍成立的设定
2. 优先把用户本轮输入映射进多个关键维度,而不是只更新一个字段
3. replyText 要明确体现“你已经理解了哪些内容”
4. 不要突然大幅改写已经成形的世界
5. 如果用户这一轮给了多条有效信息replyText 应先把这些信息自然串起来,再决定下一步
6. 可以适度替用户整理,但不要把回复写成总结报告
7. 默认继续往前推一步,不要在还没必要时突然收束或突然跳到成稿感
用户体验要求:
1. 让用户感到“我刚说的内容都被接住了”
2. 回复里可以带一点顺势整理感,但不要太像会议纪要
3. 不要无视用户刚提供的高价值细节
4. 不要让用户觉得系统在自顾自重写世界"#
}
PromptConversationMode::Compress => {
r#"当前模式compress
目标:
1. 开始收束当前设定
2. 减少无效发散
3. 让 progress 更接近可进入下一阶段
本轮行为要求:
1. 新的设定结构优先保留稳定内容,不要无端重写
2. 对用户本轮输入做高密度吸收
3. replyText 要更聚焦,不要绕圈
4. 默认只推进当前最影响 completion 的一步
5. 如果用户还在补细节,优先把细节挂回现有骨架,而不是继续开新分支
6. 可以适度提醒“还差哪类关键空位”,但不要把回复写成 checklist
7. 如果已有信息足够replyText 可以更像“确认并收束”,少一点继续发散式追问
用户体验要求:
1. 让用户感觉世界正在变得更稳,而不是越来越散
2. 让推进感更明确,但不要显得催促
3. 回复语气应更笃定一些,减少反复横跳
4. 不要把用户刚补进来的细节又冲淡掉"#
}
PromptConversationMode::RepairDirection => {
r#"当前模式repair_direction
目标:
1. 处理用户对既有设定的修正
2. 避免世界方向飘散或自相矛盾
本轮行为要求:
1. 如果用户明确改口,新的设定结构必须体现修正后的方向
2. 对已经不再成立的旧设定,不要机械保留
3. progressPercent 可以停滞,也可以小幅回落,但不能为负
4. replyText 要承认用户的修正,并顺着修正后的方向继续聊
5. 先处理“改掉什么”,再决定“往哪里继续推”
6. 不要一边口头承认用户修正,一边在设定结构里偷偷留住旧方向
7. 如果修正幅度很大replyText 可以帮助用户确认新方向已经接管当前语境
用户体验要求:
1. 让用户感到“我刚刚的纠偏真的生效了”
2. 不要和用户辩论旧方案为什么也行
3. 不要表现出对修正的不情愿
4. 回复要体现重心已经切到新方向,而不是停留在旧世界观惯性里"#
}
PromptConversationMode::ForceComplete => {
r#"当前模式force_complete
目标:
1. 基于当前方向直接补齐剩余设定
2. 生成一版尽量完整、可进入下一阶段的设定结构
3. 结束当前收集阶段
本轮行为要求:
1. 尽量保留已经形成的世界方向
2. 对明显缺失的关键维度进行合理补全
3. 不要继续拉长聊天,不要再追问用户
4. progressPercent 直接输出为 100
5. replyText 要自然引导用户点击“生成游戏设定草稿”
6. 补全时要优先做“顺着已有方向补齐”,而不是突然换题材、换气质、换主冲突
7. 可以让结果更完整,但不要补得过满、过死、过像定稿圣经
8. replyText 更像阶段完成提示,不再像继续采集信息的对话
用户体验要求:
1. 让用户感到“系统已经帮我把能补的补好了”
2. 不要在这一步突然冒出很多陌生设定把用户吓出戏
3. 回复要有完成感,但不要太官话
4. 清楚告诉用户下一步可以做什么"#
}
PromptConversationMode::Closing => {
r#"当前模式closing
目标:
1. 尽量形成一版可用的设定底子
2. 不再继续发散新世界观
本轮行为要求:
1. 优先收束,而不是扩写
2. 不要大改已经成形的核心设定
3. progressPercent 接近完成时replyText 要更像确认与推进
4. 如果用户没有大改方向,尽量让下一版内容更稳定
5. 可以轻微补足缺口,但不要再大开新支线
6. replyText 应减少探索式措辞,增加“已经基本成形”的稳定感
7. 如果只差少量空位,优先把这些空位自然补平,而不是重新打开大话题
用户体验要求:
1. 让用户感觉作品已经快成了,而不是还在无穷试探
2. 回复可以更像确认和轻推,不要继续像前期那样频繁试探
3. 保持留白感,不要把所有东西都一次说死
4. 让用户自然过渡到下一阶段,而不是突然被切断对话"#
}
}
}
fn user_signal_rules(signal: PromptUserInputSignal) -> &'static str {
match signal {
PromptUserInputSignal::Rich => {
r#"本轮用户输入信息密度高。
请尽量从这一轮里提取多个锚点,不要只更新单一方向。
如果一条输入同时影响世界方向、冲突和关系,请在新的完整设定结构中一起体现。"#
}
PromptUserInputSignal::Normal => {
r#"本轮用户输入为正常补充。
请优先顺着当前方向稳定更新,不要主动扩写太多新设定。"#
}
PromptUserInputSignal::Sparse => {
r#"本轮用户输入较少或较虚。
请保留上一版中仍然成立的内容,不要为了凑完整度而强行发明过多新设定。
replyText 要让用户容易继续往下说。"#
}
PromptUserInputSignal::Correction => {
r#"本轮用户在修正或推翻旧设定。
请优先吸收修正,不要机械复读旧版本。
新的完整设定结构必须以修正后的方向为准。"#
}
PromptUserInputSignal::Delegate => {
r#"本轮用户把部分决定权交给你。
你可以在 replyText 中给出有限度的建议,但不要突然补满整套设定。
新的完整设定结构仍应尽量建立在已有世界方向上,而不是完全重做。"#
}
}
}
fn latest_user_text(chat_history: &[JsonValue]) -> String {
chat_history
.iter()
@@ -2075,7 +1573,7 @@ fn serialize_json(value: &JsonValue, fallback: &str) -> String {
}
impl PromptUserInputSignal {
fn as_str(self) -> &'static str {
pub(crate) fn as_str(self) -> &'static str {
match self {
Self::Rich => "rich",
Self::Normal => "normal",
@@ -2087,7 +1585,7 @@ impl PromptUserInputSignal {
}
impl PromptDriftRisk {
fn as_str(self) -> &'static str {
pub(crate) fn as_str(self) -> &'static str {
match self {
Self::Low => "low",
Self::Medium => "medium",
@@ -2097,7 +1595,7 @@ impl PromptDriftRisk {
}
impl PromptConversationMode {
fn as_str(self) -> &'static str {
pub(crate) fn as_str(self) -> &'static str {
match self {
Self::Bootstrap => "bootstrap",
Self::Expand => "expand",
@@ -2111,7 +1609,7 @@ impl PromptConversationMode {
#[cfg(test)]
mod tests {
use super::extract_reply_text_from_partial_json;
use crate::custom_world_rpg_draft_prompts::extract_reply_text_from_partial_json;
#[test]
fn extract_reply_text_from_partial_json_preserves_chinese_characters() {

View File

@@ -27,8 +27,15 @@ use tokio::time::sleep;
use webp::Encoder as WebpEncoder;
use crate::{
api_response::json_success_body, auth::AuthenticatedAccessToken, http_error::AppError,
request_context::RequestContext, state::AppState,
api_response::json_success_body,
auth::AuthenticatedAccessToken,
custom_world_result_prompts::{
build_result_entity_system_prompt, build_result_entity_user_prompt,
build_result_scene_npc_system_prompt, build_result_scene_npc_user_prompt,
},
http_error::AppError,
request_context::RequestContext,
state::AppState,
};
#[derive(Clone, Debug, Deserialize)]
@@ -883,18 +890,8 @@ async fn generate_entity_with_fallback(state: &AppState, profile: &Value, kind:
return fallback;
};
let request = LlmTextRequest::new(vec![
LlmMessage::system(
"你是 RPG 自定义世界实体生成器。只输出一个 JSON 对象,不要输出 Markdown。",
),
LlmMessage::user(
json!({
"task": "generate_custom_world_entity",
"kind": kind,
"profile": profile,
"fallback": fallback,
})
.to_string(),
),
LlmMessage::system(build_result_entity_system_prompt()),
LlmMessage::user(build_result_entity_user_prompt(profile, kind, &fallback)),
]);
llm_client
@@ -915,18 +912,12 @@ async fn generate_scene_npc_with_fallback(
return fallback;
};
let request = LlmTextRequest::new(vec![
LlmMessage::system(
"你是 RPG 自定义世界场景 NPC 生成器。只输出一个 JSON 对象,不要输出 Markdown。",
),
LlmMessage::user(
json!({
"task": "generate_custom_world_scene_npc",
"landmarkId": landmark_id,
"profile": profile,
"fallback": fallback,
})
.to_string(),
),
LlmMessage::system(build_result_scene_npc_system_prompt()),
LlmMessage::user(build_result_scene_npc_user_prompt(
profile,
landmark_id,
&fallback,
)),
]);
llm_client

View File

@@ -0,0 +1,356 @@
use crate::character_animation_assets::find_motion_template;
use shared_contracts::assets::CharacterAnimationStrategy;
/// 自定义世界角色主图提示词脚本。
pub(crate) fn build_character_visual_prompt(
prompt_text: &str,
character_brief_text: Option<&str>,
) -> String {
let merged = [character_brief_text.unwrap_or_default(), prompt_text]
.into_iter()
.map(str::trim)
.filter(|value| !value.is_empty())
.collect::<Vec<_>>()
.join("\n");
format!(
"{}\n单人全身右向斜侧身3 到 4 头身,像素动作角色,纯绿色背景,服装完整,轮廓清晰,不要复杂背景。",
if merged.is_empty() {
"自定义世界角色,服装完整,姿态自然。"
} else {
merged.as_str()
}
)
}
/// 自定义世界角色主图负面提示词脚本。
pub(crate) fn build_character_visual_negative_prompt() -> String {
[
"正面视角",
"左朝向",
"完全 90 度纯右视图",
"镜头透视",
"半身像",
"脚被裁切",
"头顶被裁切",
"多角色",
"复杂背景",
"建筑场景",
"漂浮物",
"烟雾环境",
"武器消失",
"武器换手",
"额外手臂",
"额外腿",
"服装变化",
"脸部变化",
"模糊",
"运动模糊",
"文字",
"水印",
"UI 元素",
"软萌 Q版大头贴",
"儿童绘本风",
"厚涂插画感",
"低对比柔边",
]
.join("")
}
pub(crate) fn build_character_animation_prompt(
strategy: &CharacterAnimationStrategy,
prompt_text: &str,
character_brief_text: Option<&str>,
action_template_id: Option<&str>,
animation: &str,
frame_count: u32,
fps: u32,
duration_seconds: u32,
loop_: bool,
use_chroma_key: bool,
) -> String {
match strategy {
CharacterAnimationStrategy::ImageToVideo => build_ark_character_animation_prompt(
animation,
prompt_text,
character_brief_text,
action_template_id,
loop_,
use_chroma_key,
),
CharacterAnimationStrategy::ImageSequence => {
build_image_sequence_prompt(animation, prompt_text, frame_count, use_chroma_key)
}
CharacterAnimationStrategy::MotionTransfer
| CharacterAnimationStrategy::ReferenceToVideo => build_npc_animation_prompt(
animation,
prompt_text,
character_brief_text,
action_template_id,
loop_,
use_chroma_key,
fps,
duration_seconds,
),
}
}
fn build_image_sequence_prompt(
animation: &str,
prompt_text: &str,
frame_count: u32,
use_chroma_key: bool,
) -> String {
[
format!(
"同一角色连续 {} 帧动作序列,动作主题是 {}",
frame_count, animation
),
"固定机位,单人,全身,侧身朝右,保持同一套服装、发型、武器和体型。".to_string(),
"帧间动作连续,姿态逐步推进,不要换人,不要跳变,不要多余物体。".to_string(),
if use_chroma_key {
"纯绿色背景,无地面装饰,方便后期抠像。".to_string()
} else {
"背景尽量纯净,避免复杂场景。".to_string()
},
prompt_text.trim().to_string(),
]
.into_iter()
.filter(|value| !value.trim().is_empty())
.collect::<Vec<_>>()
.join(" ")
}
fn build_npc_animation_prompt(
animation: &str,
prompt_text: &str,
character_brief_text: Option<&str>,
action_template_id: Option<&str>,
loop_: bool,
use_chroma_key: bool,
fps: u32,
duration_seconds: u32,
) -> String {
let character_brief = build_compact_animation_character_brief(character_brief_text);
let action_detail_text = sanitize_animation_prompt_text(prompt_text, 140);
let loop_rule = if loop_ {
"这是循环动作,直接进入动作循环中段,不要开场静止站桩,不要把主参考图原样作为第一帧。"
.to_string()
} else if animation == "die" {
"这是死亡终结动作,首帧参考主图角色形象即可,尾帧停在死亡结束姿态,不要回到主图形象。"
.to_string()
} else {
"这是非循环动作,首帧和尾帧都要回到参考主图角色形象,中段完成动作变化。".to_string()
};
if let Some(template) = action_template_id.and_then(|id| find_motion_template(id)) {
return [
format!(
"单人 NPC 全身动作视频,动作主题是 {}。角色固定为同一人,右向斜侧身,镜头稳定,轮廓清晰,武器不可丢失。",
template.animation
),
if use_chroma_key {
"背景为纯绿色绿幕,无其他人物和场景元素,方便后期抠像。".to_string()
} else {
"背景简洁纯净,无复杂场景。".to_string()
},
if character_brief.is_empty() {
String::new()
} else {
format!("角色设定:{}", character_brief)
},
format!("动作补充:{}", template.prompt_suffix),
if action_detail_text.is_empty() {
String::new()
} else {
format!("动作细节:{}", action_detail_text)
},
format!("目标帧率 {} fps时长约 {} 秒。", fps.clamp(1, 60), duration_seconds.clamp(1, 8)),
loop_rule,
]
.into_iter()
.filter(|value| !value.trim().is_empty())
.collect::<Vec<_>>()
.join(" ");
}
[
format!("单人 NPC 全身动作视频,动作主题是 {}", animation),
"角色固定为同一人,侧身朝右,镜头稳定,轮廓清晰,武器不可丢失。".to_string(),
"动作连贯,避免服装、发型、面部、武器随机漂移。".to_string(),
if use_chroma_key {
"背景为纯绿色绿幕,无其他人物和场景元素,方便后期抠像。".to_string()
} else {
"背景简洁纯净,无复杂场景。".to_string()
},
if character_brief.is_empty() {
String::new()
} else {
format!("角色设定:{}", character_brief)
},
if action_detail_text.is_empty() {
String::new()
} else {
action_detail_text
},
format!(
"目标帧率 {} fps时长约 {} 秒。",
fps.clamp(1, 60),
duration_seconds.clamp(1, 8)
),
loop_rule,
]
.into_iter()
.filter(|value| !value.trim().is_empty())
.collect::<Vec<_>>()
.join(" ")
}
fn build_ark_character_animation_prompt(
animation: &str,
prompt_text: &str,
character_brief_text: Option<&str>,
action_template_id: Option<&str>,
loop_: bool,
use_chroma_key: bool,
) -> String {
let normalized_animation_name = animation.trim().replace(char::is_whitespace, "_");
let normalized_animation_name = if normalized_animation_name.is_empty() {
"idle".to_string()
} else {
normalized_animation_name
};
let character_brief = build_compact_animation_character_brief(character_brief_text);
let action_detail_text = sanitize_animation_prompt_text(prompt_text, 140);
let frame_rule = if loop_ {
"首帧严格使用图片1尾帧严格使用图片2循环动作必须自然闭环不要静止开场。".to_string()
} else {
"首帧严格使用图片1尾帧严格使用图片2中段完成完整动作变化收束干净。".to_string()
};
if let Some(template) = action_template_id.and_then(find_motion_template) {
return [
format!(
"单人 NPC 全身动作视频,动作英文名是 {}。角色固定为图片1和图片2中的同一人侧身朝右镜头稳定轮廓清晰武器不可丢失。",
normalized_animation_name
),
"动作连贯,避免服装、发型、面部、武器随机漂移,不要多角色,不要镜头切换。".to_string(),
if use_chroma_key {
"背景为纯绿色绿幕,无其他人物和场景元素,方便后期抠像。".to_string()
} else {
"背景简洁纯净,无复杂场景。".to_string()
},
if character_brief.is_empty() {
String::new()
} else {
format!("角色设定:{}", character_brief)
},
format!("动作补充:{}", template.prompt_suffix),
if action_detail_text.is_empty() {
String::new()
} else {
format!("动作细节:{}", action_detail_text)
},
frame_rule,
]
.into_iter()
.filter(|value| !value.trim().is_empty())
.collect::<Vec<_>>()
.join(" ");
}
[
format!(
"单人 NPC 全身动作视频,动作英文名是 {}",
normalized_animation_name
),
"角色固定为图片1和图片2中的同一人侧身朝右镜头稳定轮廓清晰武器不可丢失。"
.to_string(),
"动作连贯,避免服装、发型、面部、武器随机漂移,不要多角色,不要镜头切换。".to_string(),
if use_chroma_key {
"背景为纯绿色绿幕,无其他人物和场景元素,方便后期抠像。".to_string()
} else {
"背景简洁纯净,无复杂场景。".to_string()
},
if character_brief.is_empty() {
String::new()
} else {
format!("角色设定:{}", character_brief)
},
if action_detail_text.is_empty() {
String::new()
} else {
format!("动作细节:{}", action_detail_text)
},
frame_rule,
]
.into_iter()
.filter(|value| !value.trim().is_empty())
.collect::<Vec<_>>()
.join(" ")
}
pub(crate) fn build_fallback_moderation_safe_animation_prompt(
animation: &str,
loop_: bool,
use_chroma_key: bool,
) -> String {
[
format!("单人全身角色动作视频,动作主题是 {}", animation),
"角色固定为同一人,右向斜侧身,镜头稳定,轮廓清楚。".to_string(),
if loop_ {
"循环动作直接进入稳定循环,不要静止开场,不要定格首帧。".to_string()
} else {
"非循环动作首尾回到角色标准站姿,中段完成动作变化。".to_string()
},
if use_chroma_key {
"背景为纯绿色绿幕,无其他人物和场景元素。".to_string()
} else {
"背景简洁纯净。".to_string()
},
]
.join(" ")
}
fn sanitize_animation_prompt_text(value: &str, max_length: usize) -> String {
value
.replace(char::is_whitespace, " ")
.replace("血浆", "")
.replace("喷血", "")
.replace("鲜血", "")
.replace("断肢", "")
.replace("斩首", "")
.replace("裸体", "")
.replace("裸露", "")
.replace("色情", "")
.replace("性交", "")
.replace("死亡", "倒地结束")
.replace("死去", "倒地结束")
.replace("击杀", "倒地结束")
.replace("受击", "失衡")
.replace("受伤", "失衡")
.replace("砍杀", "挥击")
.replace("斩击", "挥击")
.split_whitespace()
.collect::<Vec<_>>()
.join(" ")
.chars()
.take(max_length)
.collect::<String>()
.trim()
.to_string()
}
fn build_compact_animation_character_brief(value: Option<&str>) -> String {
let normalized = sanitize_animation_prompt_text(value.unwrap_or_default(), 160);
if normalized.is_empty() {
return String::new();
}
normalized
.split(['/', '|', '\n', '', ',', '。', '', ';'])
.map(str::trim)
.filter(|item| !item.is_empty())
.take(4)
.collect::<Vec<_>>()
.join("")
}

View File

@@ -15,11 +15,25 @@ pub enum DraftFoundationPayloadError {
InvalidGeneratedDraft(String),
}
#[derive(Clone, Debug, PartialEq, Eq)]
pub struct CustomWorldFoundationDraftProgress {
pub phase_label: String,
pub phase_detail: String,
pub progress: u32,
}
pub async fn generate_custom_world_foundation_draft(
llm_client: &LlmClient,
session: &CustomWorldAgentSessionRecord,
mut on_progress: impl FnMut(CustomWorldFoundationDraftProgress) + Send,
) -> Result<CustomWorldFoundationDraftResult, String> {
let setting_text = build_foundation_generation_seed_text(session);
emit_foundation_draft_progress(
&mut on_progress,
"整理世界骨架",
"正在根据创作者锚点生成第一版世界框架。",
12,
);
let mut framework = request_foundation_json_stage(
llm_client,
build_custom_world_framework_prompt(setting_text.as_str()),
@@ -36,6 +50,8 @@ pub async fn generate_custom_world_foundation_draft(
&framework,
"playable",
FOUNDATION_DRAFT_PLAYABLE_COUNT,
(16, 30),
&mut on_progress,
)
.await?;
framework["playableNpcs"] = JsonValue::Array(playable_outlines.clone());
@@ -45,6 +61,8 @@ pub async fn generate_custom_world_foundation_draft(
&framework,
"story",
FOUNDATION_DRAFT_STORY_COUNT,
(30, 44),
&mut on_progress,
)
.await?;
framework["storyNpcs"] = JsonValue::Array(story_outlines.clone());
@@ -53,6 +71,8 @@ pub async fn generate_custom_world_foundation_draft(
llm_client,
&framework,
FOUNDATION_DRAFT_LANDMARK_COUNT,
(44, 56),
&mut on_progress,
)
.await?;
framework["landmarks"] = JsonValue::Array(landmark_seeds.clone());
@@ -62,6 +82,8 @@ pub async fn generate_custom_world_foundation_draft(
&framework,
&story_outlines,
&landmark_seeds,
(56, 66),
&mut on_progress,
)
.await?;
framework["landmarks"] = JsonValue::Array(landmarks.clone());
@@ -72,6 +94,8 @@ pub async fn generate_custom_world_foundation_draft(
"playable",
&playable_outlines,
"narrative",
(66, 76),
&mut on_progress,
)
.await?;
let playable_detailed = expand_foundation_role_entries(
@@ -80,6 +104,8 @@ pub async fn generate_custom_world_foundation_draft(
"playable",
&playable_narrative,
"dossier",
(76, 84),
&mut on_progress,
)
.await?;
let story_narrative = expand_foundation_role_entries(
@@ -88,6 +114,8 @@ pub async fn generate_custom_world_foundation_draft(
"story",
&story_outlines,
"narrative",
(84, 92),
&mut on_progress,
)
.await?;
let story_detailed = expand_foundation_role_entries(
@@ -96,9 +124,18 @@ pub async fn generate_custom_world_foundation_draft(
"story",
&story_narrative,
"dossier",
(92, 96),
&mut on_progress,
)
.await?;
emit_foundation_draft_progress(
&mut on_progress,
"编译世界底稿",
"正在把分批生成结果直接整理成第一版 foundation draft并同步兼容结果快照。",
97,
);
let draft_profile = build_foundation_draft_profile_from_framework(
framework,
playable_detailed,
@@ -166,6 +203,8 @@ async fn generate_foundation_role_outline_entries(
framework: &JsonValue,
role_type: &str,
total_count: usize,
progress_range: (u32, u32),
on_progress: &mut (impl FnMut(CustomWorldFoundationDraftProgress) + Send),
) -> Result<Vec<JsonValue>, String> {
let mut merged_entries = Vec::new();
let planned_batch_count = total_count
@@ -178,6 +217,24 @@ async fn generate_foundation_role_outline_entries(
let batch_count =
(total_count - merged_entries.len()).min(FOUNDATION_ROLE_OUTLINE_BATCH_SIZE);
let forbidden_names = names_from_entries(&merged_entries);
let role_label = if role_type == "playable" {
"可扮演角色"
} else {
"场景角色"
};
emit_foundation_draft_progress(
on_progress,
format!("生成{role_label}").as_str(),
format!(
"正在生成{role_label}{} / {} 批,当前已完成 {}/{}",
batch_index + 1,
planned_batch_count,
merged_entries.len(),
total_count,
)
.as_str(),
to_batch_progress(progress_range, merged_entries.len(), total_count),
);
let raw = request_foundation_json_stage(
llm_client,
build_custom_world_role_outline_batch_prompt(
@@ -210,13 +267,27 @@ async fn generate_foundation_role_outline_entries(
let key = role_key(role_type);
merged_entries.extend(array_field(&raw, key).into_iter().take(batch_count));
}
Ok(merged_entries.into_iter().take(total_count).collect())
let merged_entries: Vec<JsonValue> = merged_entries.into_iter().take(total_count).collect();
let role_label = if role_type == "playable" {
"可扮演角色"
} else {
"场景角色"
};
emit_foundation_draft_progress(
on_progress,
format!("生成{role_label}").as_str(),
format!("{role_label}已经整理完成,共 {} 个。", merged_entries.len()).as_str(),
progress_range.1,
);
Ok(merged_entries)
}
async fn generate_foundation_landmark_seed_entries(
llm_client: &LlmClient,
framework: &JsonValue,
total_count: usize,
progress_range: (u32, u32),
on_progress: &mut (impl FnMut(CustomWorldFoundationDraftProgress) + Send),
) -> Result<Vec<JsonValue>, String> {
let mut merged_entries = Vec::new();
let planned_batch_count = total_count.div_ceil(FOUNDATION_LANDMARK_BATCH_SIZE).max(1);
@@ -226,6 +297,19 @@ async fn generate_foundation_landmark_seed_entries(
}
let batch_count = (total_count - merged_entries.len()).min(FOUNDATION_LANDMARK_BATCH_SIZE);
let forbidden_names = names_from_entries(&merged_entries);
emit_foundation_draft_progress(
on_progress,
"生成关键场景",
format!(
"正在生成关键场景第 {} / {} 批,当前已完成 {}/{}",
batch_index + 1,
planned_batch_count,
merged_entries.len(),
total_count,
)
.as_str(),
to_batch_progress(progress_range, merged_entries.len(), total_count),
);
let raw = request_foundation_json_stage(
llm_client,
build_custom_world_landmark_seed_batch_prompt(framework, batch_count, &forbidden_names),
@@ -247,7 +331,14 @@ async fn generate_foundation_landmark_seed_entries(
.await?;
merged_entries.extend(array_field(&raw, "landmarks").into_iter().take(batch_count));
}
Ok(merged_entries.into_iter().take(total_count).collect())
let merged_entries: Vec<JsonValue> = merged_entries.into_iter().take(total_count).collect();
emit_foundation_draft_progress(
on_progress,
"生成关键场景",
format!("关键场景骨架已整理完成,共 {} 个。", merged_entries.len()).as_str(),
progress_range.1,
);
Ok(merged_entries)
}
async fn expand_foundation_landmark_network_entries(
@@ -255,12 +346,28 @@ async fn expand_foundation_landmark_network_entries(
framework: &JsonValue,
story_npcs: &[JsonValue],
base_entries: &[JsonValue],
progress_range: (u32, u32),
on_progress: &mut (impl FnMut(CustomWorldFoundationDraftProgress) + Send),
) -> Result<Vec<JsonValue>, String> {
let mut merged_entries = Vec::new();
for (batch_index, batch) in base_entries
let batches: Vec<&[JsonValue]> = base_entries
.chunks(FOUNDATION_LANDMARK_BATCH_SIZE)
.enumerate()
{
.collect();
let mut processed_count = 0usize;
for (batch_index, batch) in batches.iter().enumerate() {
emit_foundation_draft_progress(
on_progress,
"建立场景连接",
format!(
"正在补全场景连接第 {} / {} 批,当前已完成 {}/{}",
batch_index + 1,
batches.len(),
processed_count,
base_entries.len(),
)
.as_str(),
to_batch_progress(progress_range, processed_count, base_entries.len()),
);
let raw = request_foundation_json_stage(
llm_client,
build_custom_world_landmark_network_batch_prompt(framework, story_npcs, batch),
@@ -284,7 +391,16 @@ async fn expand_foundation_landmark_network_entries(
)
.await?;
merged_entries.extend(array_field(&raw, "landmarks"));
processed_count = processed_count
.saturating_add(batch.len())
.min(base_entries.len());
}
emit_foundation_draft_progress(
on_progress,
"建立场景连接",
"关键场景的角色分布与路径连接已经整理完成。",
progress_range.1,
);
Ok(merge_entries_by_name(base_entries, &merged_entries))
}
@@ -294,13 +410,39 @@ async fn expand_foundation_role_entries(
role_type: &str,
base_entries: &[JsonValue],
stage: &str,
progress_range: (u32, u32),
on_progress: &mut (impl FnMut(CustomWorldFoundationDraftProgress) + Send),
) -> Result<Vec<JsonValue>, String> {
let mut merged_entries = Vec::new();
for (batch_index, batch) in base_entries
let batches: Vec<&[JsonValue]> = base_entries
.chunks(FOUNDATION_ROLE_DETAIL_BATCH_SIZE)
.enumerate()
{
.collect();
let mut processed_count = 0usize;
for (batch_index, batch) in batches.iter().enumerate() {
let expected_names = names_from_entries(batch);
let role_label = if role_type == "playable" {
"可扮演角色"
} else {
"场景角色"
};
let stage_label = if stage == "narrative" {
"叙事基础"
} else {
"档案细节"
};
emit_foundation_draft_progress(
on_progress,
format!("补全{role_label}{stage_label}").as_str(),
format!(
"正在补全{role_label}{stage_label}{} / {} 批,当前已完成 {}/{}",
batch_index + 1,
batches.len(),
processed_count,
base_entries.len(),
)
.as_str(),
to_batch_progress(progress_range, processed_count, base_entries.len()),
);
let raw = request_foundation_json_stage(
llm_client,
build_custom_world_role_batch_prompt(framework, role_type, batch, stage),
@@ -326,9 +468,51 @@ async fn expand_foundation_role_entries(
)
.await?;
merged_entries.extend(array_field(&raw, role_key(role_type)));
processed_count = processed_count
.saturating_add(batch.len())
.min(base_entries.len());
}
let role_label = if role_type == "playable" {
"可扮演角色"
} else {
"场景角色"
};
let stage_label = if stage == "narrative" {
"叙事基础"
} else {
"档案细节"
};
emit_foundation_draft_progress(
on_progress,
format!("补全{role_label}{stage_label}").as_str(),
format!("{role_label}{stage_label}已经整理完成。").as_str(),
progress_range.1,
);
Ok(merge_entries_by_name(base_entries, &merged_entries))
}
fn emit_foundation_draft_progress(
on_progress: &mut (impl FnMut(CustomWorldFoundationDraftProgress) + Send),
phase_label: &str,
phase_detail: &str,
progress: u32,
) {
on_progress(CustomWorldFoundationDraftProgress {
phase_label: phase_label.to_string(),
phase_detail: phase_detail.to_string(),
progress: progress.min(100),
});
}
fn to_batch_progress(progress_range: (u32, u32), completed: usize, total: usize) -> u32 {
if total == 0 {
return progress_range.1;
}
let start = progress_range.0 as f64;
let end = progress_range.1 as f64;
let ratio = (completed as f64 / total as f64).clamp(0.0, 1.0);
(start + (end - start) * ratio).round().clamp(0.0, 100.0) as u32
}
// foundation draft 已经由 api-server 真实生成,落库前只负责把它注入现有 action payload。
pub fn build_draft_foundation_action_payload_json(
payload: &ExecuteCustomWorldAgentActionRequest,
@@ -1528,7 +1712,7 @@ mod tests {
let llm_client = build_test_llm_client(server_url);
let session = build_test_session();
let result = generate_custom_world_foundation_draft(&llm_client, &session)
let result = generate_custom_world_foundation_draft(&llm_client, &session, |_| {})
.await
.expect("draft generation should succeed");
let draft_profile = serde_json::from_str::<JsonValue>(&result.draft_profile_json)

View File

@@ -0,0 +1,39 @@
use serde_json::{Value, json};
/// 结果页新增可扮演角色 / 场景角色 / 场景的提示词脚本。
/// 这里只生成 LLM 可审计输入,不处理 fallback避免提示词规则和业务兜底混在一起。
pub(crate) fn build_result_entity_system_prompt() -> &'static str {
"你是 RPG 自定义世界实体生成器。只输出一个 JSON 对象,不要输出 Markdown。"
}
pub(crate) fn build_result_entity_user_prompt(
profile: &Value,
kind: &str,
fallback: &Value,
) -> String {
json!({
"task": "generate_custom_world_entity",
"kind": kind,
"profile": profile,
"fallback": fallback,
})
.to_string()
}
pub(crate) fn build_result_scene_npc_system_prompt() -> &'static str {
"你是 RPG 自定义世界场景 NPC 生成器。只输出一个 JSON 对象,不要输出 Markdown。"
}
pub(crate) fn build_result_scene_npc_user_prompt(
profile: &Value,
landmark_id: &str,
fallback: &Value,
) -> String {
json!({
"task": "generate_custom_world_scene_npc",
"landmarkId": landmark_id,
"profile": profile,
"fallback": fallback,
})
.to_string()
}

View File

@@ -0,0 +1,515 @@
use crate::custom_world_agent_turn::{
EightAnchorContent, PromptConversationMode, PromptDriftRisk, PromptDynamicState,
PromptUserInputSignal,
};
use module_custom_world::empty_agent_anchor_content_json;
use serde_json::Value as JsonValue;
pub(crate) const BASE_SYSTEM_PROMPT: &str = r#"你是一个负责共创游戏世界设定的专业策划。
你正在和用户一起共创一个游戏世界。每一轮你都必须读取:
1. 当前完整设定结构
2. 用户聊天记录
然后输出:
1. 一版新的完整设定结构
2. 当前 progress 百分比
3. 一段直接回复用户的话
你必须把“新的完整设定结构”视为下一轮的唯一有效版本。
你的输出会直接覆盖上一版设定结构。
你不是在做局部 patch。
你不是在做解释报告。
你不是在给开发者写分析。
你是在同时完成:
1. 世界设定更新
2. 当前推进程度判断
3. 对用户的共创回复"#;
pub(crate) const GLOBAL_HARD_RULES: &str = r#"全局硬约束:
1. 必须输出完整的设定结构,而不是只输出变化部分。
2. 新的设定结构会直接覆盖旧内容,因此不得随意丢失仍然成立的重要信息。
3. 如果用户明确修正旧设定,必须在新的设定结构中直接体现修正结果。
4. 如果用户输入信息不足,可以保留上一版中仍然成立的内容。
5. progressPercent 最低为 0不允许为负数。
6. replyText 会直接发送给用户,因此要自然、直接、可继续聊天。
7. 不要输出额外解释,不要输出 markdown 代码块,不要输出开发备注。
8. replyText 不要写成长篇策划文,不要展开大段世界观百科。
9. replyText 默认只推进当前最关键的一步,不要同时抛出很多话题。
10. replyText 不要提及“八锚点”“锚点”“结构字段”“框架字段”等内部概念词。
11. 你输出的 JSON 必须可以被直接解析。
12. 输出字段顺序必须固定为replyText、progressPercent、nextAnchorContent。"#;
pub(crate) const QUICK_FILL_EXTRA_RULES: &str = r#"用户刚刚主动要求你自动补全剩余设定。
这表示用户接受你基于当前方向自动补完剩余设定。
本轮要求:
1. 不要再继续提问
2. 直接输出一版尽量完整的设定结构
3. progressPercent 直接输出为 100
4. replyText 要告诉用户现在可以进入“生成游戏设定草稿”"#;
pub(crate) const STATE_INFERENCE_SYSTEM_PROMPT: &str = r#"你是正式生成世界设定前的一步“创作状态识别器”。
你的职责不是直接生成新设定,而是先判断:下一轮正式生成应该用什么推进策略,尤其要判断 replyText 应该更偏确认、吸收、收束、纠偏,还是启发式提问。
你必须综合以下信息判断:
1. 当前轮次 currentTurn
2. 当前完成度 progressPercent
3. 用户是否要求自动补全 quickFillRequested
4. 当前完整设定结构
5. 最近聊天记录,尤其是最近 1 到 3 轮用户消息
你需要输出 4 个字段:
1. userInputSignal只能是 rich / normal / sparse / correction / delegate
2. driftRisk只能是 low / medium / high
3. conversationMode只能是 bootstrap / expand / compress / repair_direction / force_complete / closing
4. judgementSummary1 到 2 句中文,概括你为什么这样判断,以及正式生成时最该注意什么
请按下面的语义判断。
一、userInputSignal 定义
1. rich
- 用户这一轮给了多条可直接落地的有效信息
- 这些信息可能同时覆盖世界方向、玩家处境、开局事件、冲突、关系、标志元素中的多个
- 正式生成时应优先高密度吸收,不要只更新一个点
2. normal
- 用户在顺着当前方向做正常补充
- 信息量中等,有明确新增内容,但没有明显推翻旧方向,也没有把决定权交给系统
- 正式生成时应稳定推进并自然接住用户内容
3. sparse
- 用户输入很短、很虚、很笼统,或几乎没有新增有效事实
- 例如只有一个题材词、一个气质词、一句很概括的话、一个很短的倾向表达
- 这种情况下,正式生成阶段的 replyText 应优先采用启发式提问
- 启发式提问的要求是:只问一个最容易回答、最能推动落地设计的问题
4. correction
- 用户这轮核心动作是在修正、替换、推翻、重定向旧设定
- 即使文字不长,只要主意图是“之前那个不对,现在改成这个”,也应优先判为 correction
- correction 的优先级高于 rich 和 normal
5. delegate
- 用户把部分决定权交给系统
- 例如“你来定”“你帮我补”“按你觉得合理的来”“先给我一个默认方案”
- delegate 关注的是授权关系,不只是信息多寡
二、driftRisk 定义
1. low
- 当前轮输入与已有方向基本一致
- 没有明显改口或冲突
2. medium
- 当前轮带来一定方向变化或扩张
- 还没有明显推翻旧方向,但如果处理不好,容易让设定开始发散
3. high
- 用户明确纠偏、改口、替换方向,或最近多轮反复修正
- 这时最重要的是防止旧方向重新回流到正式生成结果里
三、conversationMode 选择原则
1. bootstrap
- 适用于前期、信息少、核心方向未稳定
- replyText 更适合低压力确认和单点启发
2. expand
- 适用于方向已成形,正在顺着现有路线继续补充
- replyText 更适合总结已接住的内容并往前推一步
3. compress
- 适用于中后段,已有骨架,需要开始收束
- replyText 更适合聚焦最关键缺口,而不是继续开支线
4. repair_direction
- 适用于用户正在纠偏
- replyText 更适合先承认修正,再沿修正后的方向继续推进
5. force_complete
- 适用于用户明确要求自动补全
- replyText 不再提问,而应给出完成感和下一步引导
6. closing
- 适用于接近完成但并非强制一键补全
- replyText 更像确认与收束,而不是前期式探索
四、优先级规则
1. 如果 quickFillRequested 为 trueconversationMode 必须优先判为 force_complete
2. 如果用户核心意图是修正旧方向userInputSignal 优先判为 correctionconversationMode 通常优先考虑 repair_direction
3. 如果用户核心意图是授权系统替他补完userInputSignal 优先判为 delegate
4. 只有在没有明显纠偏、也没有明确自动补全要求时,才主要依据 currentTurn、progressPercent 和信息密度,在 bootstrap / expand / compress / closing 之间选择
五、关于 replyText 风格的专门判断要求
1. 如果用户输入较少、较虚或不够落地,正式生成阶段的 replyText 应采用启发式提问
2. 启发式提问一次最多只能提 1 个问题,不能连问两个或更多
3. 启发式提问必须问“最能推动当前设计落地”的那个问题,而不是泛泛而谈
4. 如果用户输入已经足够 rich就不要再机械提问优先吸收和推进
5. 如果用户在 correction 或 delegate 状态下replyText 是否提问要服从更高目标:纠偏生效或代为补全,不要机械套 sparse 的问法
六、关于 replyText 用语的硬约束
1. replyText 禁止提及内部结构名、锚点名、字段名、schema 名、框架词
2. 禁止出现这类内部表达:世界承诺、玩家幻想、主题边界、玩家入口、核心冲突、关键关系、隐藏线、标志元素、字段、结构、模块、八锚点
3. replyText 只能用通俗、直接、面向创作沟通的语言回应用户
4. replyText 应该围绕用户正在讨论的具体内容来落地,比如身份、开场处境、冲突、人物关系、地点、规则、气质,而不是抽象谈结构
5. judgementSummary 可以简洁提到“这轮更适合启发式提问”或“这轮应优先吸收修正”,但也不要堆内部术语
七、关于 judgementSummary 的写法
1. 必须简洁,不要写成长篇分析
2. 必须直接服务于下一轮正式生成
3. 最好同时包含两层信息:
- 为什么这么判断
- 正式生成时最该优先做什么,或最该避免什么
八、硬性约束
1. 只能输出 JSON不能输出解释、代码块或额外说明
2. 不能发明上下文里不存在的设定事实
3. 你的任务是“判断生成策略”,不是“代替正式生成直接写新设定”
4. 即使信息不完全,也必须在给定枚举里选出最合理的一组状态
5. judgementSummary 必须是中文
6. 输出值必须严格落在给定枚举中"#;
pub(crate) const STATE_INFERENCE_OUTPUT_CONTRACT: &str = r#"请严格按以下 JSON 结构输出,不要输出其他文字:
{
"userInputSignal": "normal",
"driftRisk": "low",
"conversationMode": "expand",
"judgementSummary": ""
}"#;
pub(crate) const OUTPUT_CONTRACT_REMINDER: &str = r#"请严格按以下 JSON 结构输出,不要输出其他文字:
{
"replyText": "",
"progressPercent": 0,
"nextAnchorContent": {
"worldPromise": {
"hook": "",
"differentiator": "",
"desiredExperience": ""
},
"playerFantasy": {
"playerRole": "",
"corePursuit": "",
"fearOfLoss": ""
},
"themeBoundary": {
"toneKeywords": [],
"aestheticDirectives": [],
"forbiddenDirectives": []
},
"playerEntryPoint": {
"openingIdentity": "",
"openingProblem": "",
"entryMotivation": ""
},
"coreConflict": {
"surfaceConflicts": [],
"hiddenCrisis": "",
"firstTouchedConflict": ""
},
"keyRelationships": [
{
"pairs": "",
"relationshipType": "",
"secretOrCost": ""
}
],
"hiddenLines": {
"hiddenTruths": [],
"misdirectionHints": [],
"revealPacing": ""
},
"iconicElements": {
"iconicMotifs": [],
"institutionsOrArtifacts": [],
"hardRules": []
}
}
}"#;
pub(crate) fn render_dynamic_state_context(dynamic_state: &PromptDynamicState) -> String {
format!(
"上一轮预判得到的创作状态如下。\n正式生成时必须把它作为本轮策略输入直接执行,不要重新另起一套判断。\n\n创作状态:\n- userInputSignal: {}\n- driftRisk: {}\n- conversationMode: {}\n- judgementSummary: {}",
dynamic_state.user_input_signal.as_str(),
dynamic_state.drift_risk.as_str(),
dynamic_state.conversation_mode.as_str(),
dynamic_state.judgement_summary
)
}
pub(crate) fn render_current_anchor_context(anchor_content: &EightAnchorContent) -> String {
format!(
"当前完整设定结构如下。\n你必须把它视为上一版有效世界底子。\n\n如果用户没有否定其中某部分内容,且该部分仍然成立,可以继续保留。\n如果用户明确修正了某部分内容,新的完整设定结构必须体现修正后的版本。\n\n当前完整设定结构:\n{}",
serde_json::to_string_pretty(anchor_content)
.unwrap_or_else(|_| empty_agent_anchor_content_json())
)
}
pub(crate) fn render_chat_history_context(chat_history: &[JsonValue]) -> String {
format!(
"以下是用户聊天记录。\n请重点理解最近几轮里用户新增、修正、强调的设定信息。\n不要把早期已经被用户否定的内容继续当成最终结论。\n\n用户聊天记录:\n{}",
serde_json::to_string_pretty(chat_history).unwrap_or_else(|_| "[]".to_string())
)
}
pub(crate) fn parse_json_response_text(text: &str) -> Result<JsonValue, serde_json::Error> {
let trimmed = text.trim();
if let Some(start) = trimmed.find('{')
&& let Some(end) = trimmed.rfind('}')
&& end > start
{
return serde_json::from_str::<JsonValue>(&trimmed[start..=end]);
}
serde_json::from_str::<JsonValue>(trimmed)
}
pub(crate) fn extract_reply_text_from_partial_json(text: &str) -> Option<String> {
let key_index = text.find("\"replyText\"")?;
let colon_index = text[key_index..].find(':')? + key_index;
let mut cursor = colon_index + 1;
while cursor < text.len() && text.as_bytes()[cursor].is_ascii_whitespace() {
cursor += 1;
}
if text.as_bytes().get(cursor).copied() != Some(b'"') {
return None;
}
cursor += 1;
let mut decoded = String::new();
let remainder = text.get(cursor..)?;
let mut characters = remainder.chars().peekable();
while let Some(current) = characters.next() {
if current == '"' {
return Some(decoded);
}
if current == '\\' {
let escaped = characters.next()?;
match escaped {
'"' => decoded.push('"'),
'\\' => decoded.push('\\'),
'/' => decoded.push('/'),
'b' => decoded.push('\u{0008}'),
'f' => decoded.push('\u{000C}'),
'n' => decoded.push('\n'),
'r' => decoded.push('\r'),
't' => decoded.push('\t'),
'u' => {
let mut hex = String::new();
for _ in 0..4 {
hex.push(characters.next()?);
}
if let Ok(code) = u16::from_str_radix(hex.as_str(), 16)
&& let Some(character) = char::from_u32(code as u32)
{
decoded.push(character);
}
}
other => decoded.push(other),
}
continue;
}
decoded.push(current);
}
Some(decoded)
}
pub(crate) fn parse_user_input_signal(value: Option<&JsonValue>) -> Option<PromptUserInputSignal> {
match value.and_then(JsonValue::as_str)? {
"rich" => Some(PromptUserInputSignal::Rich),
"normal" => Some(PromptUserInputSignal::Normal),
"sparse" => Some(PromptUserInputSignal::Sparse),
"correction" => Some(PromptUserInputSignal::Correction),
"delegate" => Some(PromptUserInputSignal::Delegate),
_ => None,
}
}
pub(crate) fn parse_drift_risk(value: Option<&JsonValue>) -> Option<PromptDriftRisk> {
match value.and_then(JsonValue::as_str)? {
"low" => Some(PromptDriftRisk::Low),
"medium" => Some(PromptDriftRisk::Medium),
"high" => Some(PromptDriftRisk::High),
_ => None,
}
}
pub(crate) fn parse_conversation_mode(value: Option<&JsonValue>) -> Option<PromptConversationMode> {
match value.and_then(JsonValue::as_str)? {
"bootstrap" => Some(PromptConversationMode::Bootstrap),
"expand" => Some(PromptConversationMode::Expand),
"compress" => Some(PromptConversationMode::Compress),
"repair_direction" => Some(PromptConversationMode::RepairDirection),
"force_complete" => Some(PromptConversationMode::ForceComplete),
"closing" => Some(PromptConversationMode::Closing),
_ => None,
}
}
pub(crate) fn mode_rules(mode: PromptConversationMode) -> &'static str {
match mode {
PromptConversationMode::Bootstrap => {
r#"当前模式bootstrap
目标:
1. 先把世界的基本方向抓住
2. 不要一次塞太多新设定
3. 回复要降低用户开口压力
本轮行为要求:
1. 优先从用户输入里抓世界方向、玩家视角、主题边界的线索
2. 如果用户信息很少,不要强行把整套结构一次补满
3. replyText 要像共创搭档,而不是像审问
4. 默认只推进一个最关键的问题方向
5. 如果用户刚开口,优先给“被理解感”,再轻轻推进下一步
6. 可以用一句很短的话先确认你抓到的核心方向,再提一个最好回答的问题
7. 不要把问题问得像表单采集,不要一口气追问多个维度
用户体验要求:
1. 让用户觉得“现在很容易继续往下说”
2. 不要制造被考试、被拷问、被策划问卷追着跑的感觉
3. replyText 最好短、稳、可接话
4. 如果用户信息很少,也不要显得冷淡或机械"#
}
PromptConversationMode::Expand => {
r#"当前模式expand
目标:
1. 在保持现有方向的前提下,把设定结构逐步补全
2. 尽量让一轮输入覆盖多个关键维度
本轮行为要求:
1. 继续保留上一版里仍成立的设定
2. 优先把用户本轮输入映射进多个关键维度,而不是只更新一个字段
3. replyText 要明确体现“你已经理解了哪些内容”
4. 不要突然大幅改写已经成形的世界
5. 如果用户这一轮给了多条有效信息replyText 应先把这些信息自然串起来,再决定下一步
6. 可以适度替用户整理,但不要把回复写成总结报告
7. 默认继续往前推一步,不要在还没必要时突然收束或突然跳到成稿感
用户体验要求:
1. 让用户感到“我刚说的内容都被接住了”
2. 回复里可以带一点顺势整理感,但不要太像会议纪要
3. 不要无视用户刚提供的高价值细节
4. 不要让用户觉得系统在自顾自重写世界"#
}
PromptConversationMode::Compress => {
r#"当前模式compress
目标:
1. 开始收束当前设定
2. 减少无效发散
3. 让 progress 更接近可进入下一阶段
本轮行为要求:
1. 新的设定结构优先保留稳定内容,不要无端重写
2. 对用户本轮输入做高密度吸收
3. replyText 要更聚焦,不要绕圈
4. 默认只推进当前最影响 completion 的一步
5. 如果用户还在补细节,优先把细节挂回现有骨架,而不是继续开新分支
6. 可以适度提醒“还差哪类关键空位”,但不要把回复写成 checklist
7. 如果已有信息足够replyText 可以更像“确认并收束”,少一点继续发散式追问
用户体验要求:
1. 让用户感觉世界正在变得更稳,而不是越来越散
2. 让推进感更明确,但不要显得催促
3. 回复语气应更笃定一些,减少反复横跳
4. 不要把用户刚补进来的细节又冲淡掉"#
}
PromptConversationMode::RepairDirection => {
r#"当前模式repair_direction
目标:
1. 处理用户对既有设定的修正
2. 避免世界方向飘散或自相矛盾
本轮行为要求:
1. 如果用户明确改口,新的设定结构必须体现修正后的方向
2. 对已经不再成立的旧设定,不要机械保留
3. progressPercent 可以停滞,也可以小幅回落,但不能为负
4. replyText 要承认用户的修正,并顺着修正后的方向继续聊
5. 先处理“改掉什么”,再决定“往哪里继续推”
6. 不要一边口头承认用户修正,一边在设定结构里偷偷留住旧方向
7. 如果修正幅度很大replyText 可以帮助用户确认新方向已经接管当前语境
用户体验要求:
1. 让用户感到“我刚刚的纠偏真的生效了”
2. 不要和用户辩论旧方案为什么也行
3. 不要表现出对修正的不情愿
4. 回复要体现重心已经切到新方向,而不是停留在旧世界观惯性里"#
}
PromptConversationMode::ForceComplete => {
r#"当前模式force_complete
目标:
1. 基于当前方向直接补齐剩余设定
2. 生成一版尽量完整、可进入下一阶段的设定结构
3. 结束当前收集阶段
本轮行为要求:
1. 尽量保留已经形成的世界方向
2. 对明显缺失的关键维度进行合理补全
3. 不要继续拉长聊天,不要再追问用户
4. progressPercent 直接输出为 100
5. replyText 要自然引导用户点击“生成游戏设定草稿”
6. 补全时要优先做“顺着已有方向补齐”,而不是突然换题材、换气质、换主冲突
7. 可以让结果更完整,但不要补得过满、过死、过像定稿圣经
8. replyText 更像阶段完成提示,不再像继续采集信息的对话
用户体验要求:
1. 让用户感到“系统已经帮我把能补的补好了”
2. 不要在这一步突然冒出很多陌生设定把用户吓出戏
3. 回复要有完成感,但不要太官话
4. 清楚告诉用户下一步可以做什么"#
}
PromptConversationMode::Closing => {
r#"当前模式closing
目标:
1. 尽量形成一版可用的设定底子
2. 不再继续发散新世界观
本轮行为要求:
1. 优先收束,而不是扩写
2. 不要大改已经成形的核心设定
3. progressPercent 接近完成时replyText 要更像确认与推进
4. 如果用户没有大改方向,尽量让下一版内容更稳定
5. 可以轻微补足缺口,但不要再大开新支线
6. replyText 应减少探索式措辞,增加“已经基本成形”的稳定感
7. 如果只差少量空位,优先把这些空位自然补平,而不是重新打开大话题
用户体验要求:
1. 让用户感觉作品已经快成了,而不是还在无穷试探
2. 回复可以更像确认和轻推,不要继续像前期那样频繁试探
3. 保持留白感,不要把所有东西都一次说死
4. 让用户自然过渡到下一阶段,而不是突然被切断对话"#
}
}
}
pub(crate) fn user_signal_rules(signal: PromptUserInputSignal) -> &'static str {
match signal {
PromptUserInputSignal::Rich => {
r#"本轮用户输入信息密度高。
请尽量从这一轮里提取多个锚点,不要只更新单一方向。
如果一条输入同时影响世界方向、冲突和关系,请在新的完整设定结构中一起体现。"#
}
PromptUserInputSignal::Normal => {
r#"本轮用户输入为正常补充。
请优先顺着当前方向稳定更新,不要主动扩写太多新设定。"#
}
PromptUserInputSignal::Sparse => {
r#"本轮用户输入较少或较虚。
请保留上一版中仍然成立的内容,不要为了凑完整度而强行发明过多新设定。
replyText 要让用户容易继续往下说。"#
}
PromptUserInputSignal::Correction => {
r#"本轮用户在修正或推翻旧设定。
请优先吸收修正,不要机械复读旧版本。
新的完整设定结构必须以修正后的方向为准。"#
}
PromptUserInputSignal::Delegate => {
r#"本轮用户把部分决定权交给你。
你可以在 replyText 中给出有限度的建议,但不要突然补满整套设定。
新的完整设定结构仍应尽量建立在已有世界方向上,而不是完全重做。"#
}
}
}

View File

@@ -19,7 +19,10 @@ mod custom_world;
mod custom_world_agent_entities;
mod custom_world_agent_turn;
mod custom_world_ai;
mod custom_world_asset_prompts;
mod custom_world_foundation_draft;
mod custom_world_result_prompts;
mod custom_world_rpg_draft_prompts;
mod error_middleware;
mod health;
mod http_error;

View File

@@ -562,6 +562,21 @@ pub struct CustomWorldAgentOperationGetInput {
pub operation_id: String,
}
#[cfg_attr(feature = "spacetime-types", derive(SpacetimeType))]
#[derive(Clone, Debug, PartialEq, Eq, Serialize, Deserialize)]
pub struct CustomWorldAgentOperationProgressInput {
pub session_id: String,
pub owner_user_id: String,
pub operation_id: String,
pub operation_type: RpgAgentOperationType,
pub operation_status: RpgAgentOperationStatus,
pub phase_label: String,
pub phase_detail: String,
pub operation_progress: u32,
pub error_message: Option<String>,
pub updated_at_micros: i64,
}
#[cfg_attr(feature = "spacetime-types", derive(SpacetimeType))]
#[derive(Clone, Debug, PartialEq, Eq, Serialize, Deserialize)]
pub struct CustomWorldAgentOperationProcedureResult {
@@ -1205,6 +1220,24 @@ pub fn validate_custom_world_agent_operation_get_input(
Ok(())
}
pub fn validate_custom_world_agent_operation_progress_input(
input: &CustomWorldAgentOperationProgressInput,
) -> Result<(), CustomWorldFieldError> {
validate_custom_world_agent_operation_get_input(&CustomWorldAgentOperationGetInput {
session_id: input.session_id.clone(),
owner_user_id: input.owner_user_id.clone(),
operation_id: input.operation_id.clone(),
})?;
validate_custom_world_agent_operation_fields(
&input.operation_id,
&input.session_id,
&input.phase_label,
input.operation_progress,
)?;
Ok(())
}
pub fn validate_custom_world_works_list_input(
input: &CustomWorldWorksListInput,
) -> Result<(), CustomWorldFieldError> {

View File

@@ -187,6 +187,7 @@ use crate::module_bindings::{
CustomWorldAgentMessageSubmitInput as BindingCustomWorldAgentMessageSubmitInput,
CustomWorldAgentOperationGetInput as BindingCustomWorldAgentOperationGetInput,
CustomWorldAgentOperationProcedureResult as BindingCustomWorldAgentOperationProcedureResult,
CustomWorldAgentOperationProgressInput as BindingCustomWorldAgentOperationProgressInput,
CustomWorldAgentOperationSnapshot as BindingCustomWorldAgentOperationSnapshot,
CustomWorldAgentSessionCreateInput as BindingCustomWorldAgentSessionCreateInput,
CustomWorldAgentSessionGetInput as BindingCustomWorldAgentSessionGetInput,
@@ -373,6 +374,7 @@ use crate::module_bindings::{
swap_puzzle_pieces_procedure::swap_puzzle_pieces as _,
unpublish_custom_world_profile_and_return_procedure::unpublish_custom_world_profile_and_return as _,
update_puzzle_work_procedure::update_puzzle_work as _,
upsert_custom_world_agent_operation_progress_procedure::upsert_custom_world_agent_operation_progress as _,
upsert_custom_world_profile_and_return_procedure::upsert_custom_world_profile_and_return as _,
upsert_platform_browse_history_and_return_procedure::upsert_platform_browse_history_and_return as _,
upsert_runtime_setting_and_return_procedure::upsert_runtime_setting_and_return as _,
@@ -1886,6 +1888,36 @@ impl SpacetimeClient {
.await
}
pub async fn upsert_custom_world_agent_operation_progress(
&self,
input: CustomWorldAgentOperationProgressRecordInput,
) -> Result<CustomWorldAgentOperationRecord, SpacetimeClientError> {
let procedure_input = BindingCustomWorldAgentOperationProgressInput {
session_id: input.session_id,
owner_user_id: input.owner_user_id,
operation_id: input.operation_id,
operation_type: map_custom_world_agent_operation_type(input.operation_type.as_str()),
operation_status: map_custom_world_agent_operation_status(input.operation_status.as_str()),
phase_label: input.phase_label,
phase_detail: input.phase_detail,
operation_progress: input.operation_progress,
error_message: input.error_message,
updated_at_micros: input.updated_at_micros,
};
self.call_after_connect(move |connection, sender| {
connection
.procedures()
.upsert_custom_world_agent_operation_progress_then(procedure_input, move |_, result| {
let mapped = result
.map_err(|error| SpacetimeClientError::Procedure(error.to_string()))
.and_then(map_custom_world_agent_operation_procedure_result);
send_once(&sender, mapped);
});
})
.await
}
pub async fn list_platform_browse_history(
&self,
user_id: String,
@@ -5274,6 +5306,24 @@ fn format_rpg_agent_operation_type(
}
}
fn map_custom_world_agent_operation_type(value: &str) -> crate::module_bindings::RpgAgentOperationType {
match value.trim() {
"draft_foundation" => crate::module_bindings::RpgAgentOperationType::DraftFoundation,
"update_draft_card" => crate::module_bindings::RpgAgentOperationType::UpdateDraftCard,
"sync_result_profile" => crate::module_bindings::RpgAgentOperationType::SyncResultProfile,
"generate_characters" => crate::module_bindings::RpgAgentOperationType::GenerateCharacters,
"generate_landmarks" => crate::module_bindings::RpgAgentOperationType::GenerateLandmarks,
"generate_role_assets" => crate::module_bindings::RpgAgentOperationType::GenerateRoleAssets,
"sync_role_assets" => crate::module_bindings::RpgAgentOperationType::SyncRoleAssets,
"generate_scene_assets" => crate::module_bindings::RpgAgentOperationType::GenerateSceneAssets,
"sync_scene_assets" => crate::module_bindings::RpgAgentOperationType::SyncSceneAssets,
"expand_long_tail" => crate::module_bindings::RpgAgentOperationType::ExpandLongTail,
"publish_world" => crate::module_bindings::RpgAgentOperationType::PublishWorld,
"revert_checkpoint" => crate::module_bindings::RpgAgentOperationType::RevertCheckpoint,
_ => crate::module_bindings::RpgAgentOperationType::ProcessMessage,
}
}
fn format_rpg_agent_operation_status(
value: crate::module_bindings::RpgAgentOperationStatus,
) -> &'static str {
@@ -5285,6 +5335,15 @@ fn format_rpg_agent_operation_status(
}
}
fn map_custom_world_agent_operation_status(value: &str) -> crate::module_bindings::RpgAgentOperationStatus {
match value.trim() {
"queued" => crate::module_bindings::RpgAgentOperationStatus::Queued,
"completed" => crate::module_bindings::RpgAgentOperationStatus::Completed,
"failed" => crate::module_bindings::RpgAgentOperationStatus::Failed,
_ => crate::module_bindings::RpgAgentOperationStatus::Running,
}
}
fn parse_rpg_agent_operation_status_record(
value: &str,
) -> Result<crate::module_bindings::RpgAgentOperationStatus, SpacetimeClientError> {
@@ -5970,6 +6029,20 @@ pub struct CustomWorldAgentOperationRecord {
pub error_message: Option<String>,
}
#[derive(Clone, Debug, PartialEq, Eq)]
pub struct CustomWorldAgentOperationProgressRecordInput {
pub session_id: String,
pub owner_user_id: String,
pub operation_id: String,
pub operation_type: String,
pub operation_status: String,
pub phase_label: String,
pub phase_detail: String,
pub operation_progress: u32,
pub error_message: Option<String>,
pub updated_at_micros: i64,
}
#[derive(Clone, Debug, PartialEq)]
pub struct CustomWorldDraftCardRecord {
pub card_id: String,

View File

@@ -0,0 +1,32 @@
// THIS FILE IS AUTOMATICALLY GENERATED BY SPACETIMEDB. EDITS TO THIS FILE
// WILL NOT BE SAVED. MODIFY TABLES IN YOUR MODULE SOURCE CODE INSTEAD.
#![allow(unused, clippy::all)]
use spacetimedb_sdk::__codegen::{
self as __sdk,
__lib,
__sats,
__ws,
};
use super::rpg_agent_operation_status_type::RpgAgentOperationStatus;
use super::rpg_agent_operation_type_type::RpgAgentOperationType;
#[derive(__lib::ser::Serialize, __lib::de::Deserialize, Clone, PartialEq, Debug)]
#[sats(crate = __lib)]
pub struct CustomWorldAgentOperationProgressInput {
pub session_id: String,
pub owner_user_id: String,
pub operation_id: String,
pub operation_type: RpgAgentOperationType,
pub operation_status: RpgAgentOperationStatus,
pub phase_label: String,
pub phase_detail: String,
pub operation_progress: u32,
pub error_message: Option::<String>,
pub updated_at_micros: i64,
}
impl __sdk::InModule for CustomWorldAgentOperationProgressInput {
type Module = super::RemoteModule;
}

View File

@@ -44,13 +44,10 @@ pub mod asset_object_access_policy_type;
pub mod asset_object_procedure_result_type;
pub mod asset_object_upsert_input_type;
pub mod asset_object_upsert_snapshot_type;
<<<<<<< HEAD
=======
pub mod auth_store_snapshot_type;
pub mod auth_store_snapshot_procedure_result_type;
pub mod auth_store_snapshot_record_type;
pub mod auth_store_snapshot_upsert_input_type;
>>>>>>> 4f272a50 ( Spacetime )
pub mod battle_mode_type;
pub mod battle_state_type;
pub mod battle_state_input_type;
@@ -115,6 +112,7 @@ pub mod custom_world_agent_message_submit_input_type;
pub mod custom_world_agent_operation_type;
pub mod custom_world_agent_operation_get_input_type;
pub mod custom_world_agent_operation_procedure_result_type;
pub mod custom_world_agent_operation_progress_input_type;
pub mod custom_world_agent_operation_snapshot_type;
pub mod custom_world_agent_session_type;
pub mod custom_world_agent_session_create_input_type;
@@ -346,10 +344,7 @@ pub mod ai_task_stage_table;
pub mod ai_text_chunk_table;
pub mod asset_entity_binding_table;
pub mod asset_object_table;
<<<<<<< HEAD
=======
pub mod auth_store_snapshot_table;
>>>>>>> 4f272a50 ( Spacetime )
pub mod battle_state_table;
pub mod big_fish_agent_message_table;
pub mod big_fish_asset_slot_table;
@@ -407,17 +402,11 @@ pub mod delete_runtime_snapshot_and_return_procedure;
pub mod drag_puzzle_piece_or_group_procedure;
pub mod execute_custom_world_agent_action_procedure;
pub mod fail_ai_task_and_return_procedure;
<<<<<<< HEAD
pub mod finalize_big_fish_agent_message_turn_procedure;
pub mod finalize_custom_world_agent_message_turn_procedure;
pub mod finalize_puzzle_agent_message_turn_procedure;
pub mod generate_big_fish_asset_procedure;
=======
pub mod finalize_custom_world_agent_message_turn_procedure;
pub mod finalize_puzzle_agent_message_turn_procedure;
pub mod generate_big_fish_asset_procedure;
pub mod get_auth_store_snapshot_procedure;
>>>>>>> 4f272a50 ( Spacetime )
pub mod get_battle_state_procedure;
pub mod get_big_fish_run_procedure;
pub mod get_big_fish_session_procedure;
@@ -471,6 +460,7 @@ pub mod swap_puzzle_pieces_procedure;
pub mod unpublish_custom_world_profile_and_return_procedure;
pub mod update_puzzle_work_procedure;
pub mod upsert_auth_store_snapshot_procedure;
pub mod upsert_custom_world_agent_operation_progress_procedure;
pub mod upsert_chapter_progression_and_return_procedure;
pub mod upsert_custom_world_profile_and_return_procedure;
pub mod upsert_npc_state_and_return_procedure;
@@ -511,13 +501,10 @@ pub use asset_object_access_policy_type::AssetObjectAccessPolicy;
pub use asset_object_procedure_result_type::AssetObjectProcedureResult;
pub use asset_object_upsert_input_type::AssetObjectUpsertInput;
pub use asset_object_upsert_snapshot_type::AssetObjectUpsertSnapshot;
<<<<<<< HEAD
=======
pub use auth_store_snapshot_type::AuthStoreSnapshot;
pub use auth_store_snapshot_procedure_result_type::AuthStoreSnapshotProcedureResult;
pub use auth_store_snapshot_record_type::AuthStoreSnapshotRecord;
pub use auth_store_snapshot_upsert_input_type::AuthStoreSnapshotUpsertInput;
>>>>>>> 4f272a50 ( Spacetime )
pub use battle_mode_type::BattleMode;
pub use battle_state_type::BattleState;
pub use battle_state_input_type::BattleStateInput;
@@ -582,6 +569,7 @@ pub use custom_world_agent_message_submit_input_type::CustomWorldAgentMessageSub
pub use custom_world_agent_operation_type::CustomWorldAgentOperation;
pub use custom_world_agent_operation_get_input_type::CustomWorldAgentOperationGetInput;
pub use custom_world_agent_operation_procedure_result_type::CustomWorldAgentOperationProcedureResult;
pub use custom_world_agent_operation_progress_input_type::CustomWorldAgentOperationProgressInput;
pub use custom_world_agent_operation_snapshot_type::CustomWorldAgentOperationSnapshot;
pub use custom_world_agent_session_type::CustomWorldAgentSession;
pub use custom_world_agent_session_create_input_type::CustomWorldAgentSessionCreateInput;
@@ -789,10 +777,7 @@ pub use ai_task_stage_table::*;
pub use ai_text_chunk_table::*;
pub use asset_entity_binding_table::*;
pub use asset_object_table::*;
<<<<<<< HEAD
=======
pub use auth_store_snapshot_table::*;
>>>>>>> 4f272a50 ( Spacetime )
pub use battle_state_table::*;
pub use big_fish_agent_message_table::*;
pub use big_fish_asset_slot_table::*;
@@ -874,17 +859,11 @@ pub use delete_runtime_snapshot_and_return_procedure::delete_runtime_snapshot_an
pub use drag_puzzle_piece_or_group_procedure::drag_puzzle_piece_or_group;
pub use execute_custom_world_agent_action_procedure::execute_custom_world_agent_action;
pub use fail_ai_task_and_return_procedure::fail_ai_task_and_return;
<<<<<<< HEAD
pub use finalize_big_fish_agent_message_turn_procedure::finalize_big_fish_agent_message_turn;
pub use finalize_custom_world_agent_message_turn_procedure::finalize_custom_world_agent_message_turn;
pub use finalize_puzzle_agent_message_turn_procedure::finalize_puzzle_agent_message_turn;
pub use generate_big_fish_asset_procedure::generate_big_fish_asset;
=======
pub use finalize_custom_world_agent_message_turn_procedure::finalize_custom_world_agent_message_turn;
pub use finalize_puzzle_agent_message_turn_procedure::finalize_puzzle_agent_message_turn;
pub use generate_big_fish_asset_procedure::generate_big_fish_asset;
pub use get_auth_store_snapshot_procedure::get_auth_store_snapshot;
>>>>>>> 4f272a50 ( Spacetime )
pub use get_battle_state_procedure::get_battle_state;
pub use get_big_fish_run_procedure::get_big_fish_run;
pub use get_big_fish_session_procedure::get_big_fish_session;
@@ -938,6 +917,7 @@ pub use swap_puzzle_pieces_procedure::swap_puzzle_pieces;
pub use unpublish_custom_world_profile_and_return_procedure::unpublish_custom_world_profile_and_return;
pub use update_puzzle_work_procedure::update_puzzle_work;
pub use upsert_auth_store_snapshot_procedure::upsert_auth_store_snapshot;
pub use upsert_custom_world_agent_operation_progress_procedure::upsert_custom_world_agent_operation_progress;
pub use upsert_chapter_progression_and_return_procedure::upsert_chapter_progression_and_return;
pub use upsert_custom_world_profile_and_return_procedure::upsert_custom_world_profile_and_return;
pub use upsert_npc_state_and_return_procedure::upsert_npc_state_and_return;
@@ -1249,10 +1229,7 @@ impl TryFrom<__ws::v2::TransactionUpdate> for DbUpdate {
"ai_text_chunk" => db_update.ai_text_chunk.append(ai_text_chunk_table::parse_table_update(table_update)?),
"asset_entity_binding" => db_update.asset_entity_binding.append(asset_entity_binding_table::parse_table_update(table_update)?),
"asset_object" => db_update.asset_object.append(asset_object_table::parse_table_update(table_update)?),
<<<<<<< HEAD
=======
"auth_store_snapshot" => db_update.auth_store_snapshot.append(auth_store_snapshot_table::parse_table_update(table_update)?),
>>>>>>> 4f272a50 ( Spacetime )
"battle_state" => db_update.battle_state.append(battle_state_table::parse_table_update(table_update)?),
"big_fish_agent_message" => db_update.big_fish_agent_message.append(big_fish_agent_message_table::parse_table_update(table_update)?),
"big_fish_asset_slot" => db_update.big_fish_asset_slot.append(big_fish_asset_slot_table::parse_table_update(table_update)?),
@@ -1313,10 +1290,7 @@ impl __sdk::DbUpdate for DbUpdate {
diff.ai_text_chunk = cache.apply_diff_to_table::<AiTextChunk>("ai_text_chunk", &self.ai_text_chunk).with_updates_by_pk(|row| &row.text_chunk_row_id);
diff.asset_entity_binding = cache.apply_diff_to_table::<AssetEntityBinding>("asset_entity_binding", &self.asset_entity_binding).with_updates_by_pk(|row| &row.binding_id);
diff.asset_object = cache.apply_diff_to_table::<AssetObject>("asset_object", &self.asset_object).with_updates_by_pk(|row| &row.asset_object_id);
<<<<<<< HEAD
=======
diff.auth_store_snapshot = cache.apply_diff_to_table::<AuthStoreSnapshot>("auth_store_snapshot", &self.auth_store_snapshot).with_updates_by_pk(|row| &row.snapshot_id);
>>>>>>> 4f272a50 ( Spacetime )
diff.battle_state = cache.apply_diff_to_table::<BattleState>("battle_state", &self.battle_state).with_updates_by_pk(|row| &row.battle_state_id);
diff.big_fish_agent_message = cache.apply_diff_to_table::<BigFishAgentMessage>("big_fish_agent_message", &self.big_fish_agent_message).with_updates_by_pk(|row| &row.message_id);
diff.big_fish_asset_slot = cache.apply_diff_to_table::<BigFishAssetSlot>("big_fish_asset_slot", &self.big_fish_asset_slot).with_updates_by_pk(|row| &row.slot_id);
@@ -1362,10 +1336,7 @@ for table_rows in raw.tables {
"ai_text_chunk" => db_update.ai_text_chunk.append(__sdk::parse_row_list_as_inserts(table_rows.rows)?),
"asset_entity_binding" => db_update.asset_entity_binding.append(__sdk::parse_row_list_as_inserts(table_rows.rows)?),
"asset_object" => db_update.asset_object.append(__sdk::parse_row_list_as_inserts(table_rows.rows)?),
<<<<<<< HEAD
=======
"auth_store_snapshot" => db_update.auth_store_snapshot.append(__sdk::parse_row_list_as_inserts(table_rows.rows)?),
>>>>>>> 4f272a50 ( Spacetime )
"battle_state" => db_update.battle_state.append(__sdk::parse_row_list_as_inserts(table_rows.rows)?),
"big_fish_agent_message" => db_update.big_fish_agent_message.append(__sdk::parse_row_list_as_inserts(table_rows.rows)?),
"big_fish_asset_slot" => db_update.big_fish_asset_slot.append(__sdk::parse_row_list_as_inserts(table_rows.rows)?),
@@ -1411,10 +1382,7 @@ for table_rows in raw.tables {
"ai_text_chunk" => db_update.ai_text_chunk.append(__sdk::parse_row_list_as_deletes(table_rows.rows)?),
"asset_entity_binding" => db_update.asset_entity_binding.append(__sdk::parse_row_list_as_deletes(table_rows.rows)?),
"asset_object" => db_update.asset_object.append(__sdk::parse_row_list_as_deletes(table_rows.rows)?),
<<<<<<< HEAD
=======
"auth_store_snapshot" => db_update.auth_store_snapshot.append(__sdk::parse_row_list_as_deletes(table_rows.rows)?),
>>>>>>> 4f272a50 ( Spacetime )
"battle_state" => db_update.battle_state.append(__sdk::parse_row_list_as_deletes(table_rows.rows)?),
"big_fish_agent_message" => db_update.big_fish_agent_message.append(__sdk::parse_row_list_as_deletes(table_rows.rows)?),
"big_fish_asset_slot" => db_update.big_fish_asset_slot.append(__sdk::parse_row_list_as_deletes(table_rows.rows)?),
@@ -1511,10 +1479,7 @@ impl<'r> __sdk::AppliedDiff<'r> for AppliedDiff<'r> {
callbacks.invoke_table_row_callbacks::<AiTextChunk>("ai_text_chunk", &self.ai_text_chunk, event);
callbacks.invoke_table_row_callbacks::<AssetEntityBinding>("asset_entity_binding", &self.asset_entity_binding, event);
callbacks.invoke_table_row_callbacks::<AssetObject>("asset_object", &self.asset_object, event);
<<<<<<< HEAD
=======
callbacks.invoke_table_row_callbacks::<AuthStoreSnapshot>("auth_store_snapshot", &self.auth_store_snapshot, event);
>>>>>>> 4f272a50 ( Spacetime )
callbacks.invoke_table_row_callbacks::<BattleState>("battle_state", &self.battle_state, event);
callbacks.invoke_table_row_callbacks::<BigFishAgentMessage>("big_fish_agent_message", &self.big_fish_agent_message, event);
callbacks.invoke_table_row_callbacks::<BigFishAssetSlot>("big_fish_asset_slot", &self.big_fish_asset_slot, event);

View File

@@ -0,0 +1,64 @@
// THIS FILE IS AUTOMATICALLY GENERATED BY SPACETIMEDB. EDITS TO THIS FILE
// WILL NOT BE SAVED. MODIFY TABLES IN YOUR MODULE SOURCE CODE INSTEAD.
#![allow(unused, clippy::all)]
use spacetimedb_sdk::__codegen::{
self as __sdk,
__lib,
__sats,
__ws,
};
use super::custom_world_agent_operation_procedure_result_type::CustomWorldAgentOperationProcedureResult;
use super::custom_world_agent_operation_progress_input_type::CustomWorldAgentOperationProgressInput;
#[derive(__lib::ser::Serialize, __lib::de::Deserialize, Clone, PartialEq, Debug)]
#[sats(crate = __lib)]
struct UpsertCustomWorldAgentOperationProgressArgs {
pub input: CustomWorldAgentOperationProgressInput,
}
impl __sdk::InModule for UpsertCustomWorldAgentOperationProgressArgs {
type Module = super::RemoteModule;
}
#[allow(non_camel_case_types)]
/// Extension trait for access to the procedure `upsert_custom_world_agent_operation_progress`.
///
/// Implemented for [`super::RemoteProcedures`].
pub trait upsert_custom_world_agent_operation_progress {
fn upsert_custom_world_agent_operation_progress(
&self,
input: CustomWorldAgentOperationProgressInput,
) {
self.upsert_custom_world_agent_operation_progress_then(input, |_, _| {});
}
fn upsert_custom_world_agent_operation_progress_then(
&self,
input: CustomWorldAgentOperationProgressInput,
__callback: impl FnOnce(
&super::ProcedureEventContext,
Result<CustomWorldAgentOperationProcedureResult, __sdk::InternalError>,
) + Send
+ 'static,
);
}
impl upsert_custom_world_agent_operation_progress for super::RemoteProcedures {
fn upsert_custom_world_agent_operation_progress_then(
&self,
input: CustomWorldAgentOperationProgressInput,
__callback: impl FnOnce(
&super::ProcedureEventContext,
Result<CustomWorldAgentOperationProcedureResult, __sdk::InternalError>,
) + Send
+ 'static,
) {
self.imp.invoke_procedure_with_callback::<_, CustomWorldAgentOperationProcedureResult>(
"upsert_custom_world_agent_operation_progress",
UpsertCustomWorldAgentOperationProgressArgs { input },
__callback,
);
}
}

View File

@@ -1588,13 +1588,15 @@ fn execute_custom_world_agent_action_tx(
}
"publish_world" => execute_publish_world_action(ctx, &session, &input, &payload),
"revert_checkpoint" => execute_revert_checkpoint_action(ctx, &session, &input, &payload),
"generate_characters"
| "generate_landmarks"
| "generate_role_assets"
| "sync_role_assets"
| "generate_scene_assets"
| "sync_scene_assets"
| "expand_long_tail" => execute_placeholder_custom_world_action(ctx, &session, &input),
"generate_characters" | "generate_landmarks" => {
execute_generate_entities_action(ctx, &session, &input, &payload)
}
"generate_role_assets" | "generate_scene_assets" => {
execute_prepare_asset_studio_action(ctx, &session, &input, &payload)
}
"sync_role_assets" => execute_sync_role_assets_action(ctx, &session, &input, &payload),
"sync_scene_assets" => execute_sync_scene_assets_action(ctx, &session, &input, &payload),
"expand_long_tail" => execute_placeholder_custom_world_action(ctx, &session, &input),
other => Err(format!("custom world action `{other}` 当前尚未支持")),
}
}
@@ -2134,6 +2136,146 @@ fn execute_revert_checkpoint_action(
Ok(build_custom_world_agent_operation_snapshot(&operation))
}
fn execute_prepare_asset_studio_action(
ctx: &ReducerContext,
session: &CustomWorldAgentSession,
input: &CustomWorldAgentActionExecuteInput,
payload: &JsonMap<String, JsonValue>,
) -> Result<CustomWorldAgentOperationSnapshot, String> {
ensure_draft_refining_stage(session.stage, input.action.as_str())?;
let draft_profile = parse_optional_session_object(session.draft_profile_json.as_deref())
.ok_or_else(|| format!("{} requires an existing draft foundation", input.action))?;
let (focus_id, operation_type, message_text, phase_label, phase_detail) =
if input.action == "generate_role_assets" {
let role_id = read_first_payload_text(payload, "roleIds", "roleId")
.ok_or_else(|| "generate_role_assets requires roleIds".to_string())?;
let role = find_profile_entity_by_id(&draft_profile, &["playableNpcs", "storyNpcs"], &role_id)
.ok_or_else(|| "未找到目标角色,无法进入角色资产工坊。".to_string())?;
let role_name = read_optional_text_field(role, &["name"]).unwrap_or_else(|| "角色".to_string());
(
role_id,
RpgAgentOperationType::GenerateRoleAssets,
format!("已为「{}」准备好角色资产工坊,先生成主图候选,再补核心动作。", role_name),
"角色资产工坊已就绪",
format!("{}」现在可以开始生成主图和动作。", role_name),
)
} else {
let scene_id = read_first_payload_text(payload, "sceneIds", "sceneId")
.ok_or_else(|| "generate_scene_assets requires sceneIds".to_string())?;
let scene_kind = payload.get("sceneKind").and_then(JsonValue::as_str).map(str::trim).unwrap_or("landmark");
let scene = if scene_kind == "camp" {
draft_profile.get("camp").and_then(JsonValue::as_object)
} else {
find_profile_entity_by_id(&draft_profile, &["landmarks"], &scene_id)
}
.ok_or_else(|| "未找到目标场景,无法进入场景资产工坊。".to_string())?;
let scene_name = read_optional_text_field(scene, &["name"])
.unwrap_or_else(|| if scene_kind == "camp" { "开局营地" } else { "未命名场景" }.to_string());
(
scene_id,
RpgAgentOperationType::GenerateSceneAssets,
format!("已为「{}」准备好场景图工坊,保存生成结果后会自动同步回当前草稿。", scene_name),
"场景资产工坊已就绪",
format!("{}」现在可以继续生成和确认正式场景图。", scene_name),
)
};
let next_session = rebuild_custom_world_agent_session_row(
session,
CustomWorldAgentSessionPatch {
stage: Some(RpgAgentStage::VisualRefining),
focus_card_id: Some(Some(focus_id)),
last_assistant_reply: Some(Some(message_text.clone())),
updated_at_micros: Some(input.submitted_at_micros),
..CustomWorldAgentSessionPatch::default()
},
)?;
replace_custom_world_agent_session(ctx, session, next_session);
append_custom_world_action_result_message(ctx, &session.session_id, &input.operation_id, &message_text, input.submitted_at_micros);
let operation = build_and_insert_custom_world_operation(ctx, &input.operation_id, &session.session_id, operation_type, phase_label, &phase_detail, input.submitted_at_micros);
Ok(build_custom_world_agent_operation_snapshot(&operation))
}
fn execute_sync_role_assets_action(
ctx: &ReducerContext,
session: &CustomWorldAgentSession,
input: &CustomWorldAgentActionExecuteInput,
payload: &JsonMap<String, JsonValue>,
) -> Result<CustomWorldAgentOperationSnapshot, String> {
ensure_draft_refining_stage(session.stage, "sync_role_assets")?;
let mut draft_profile = parse_optional_session_object(session.draft_profile_json.as_deref())
.ok_or_else(|| "sync_role_assets requires an existing draft foundation".to_string())?;
let role_id = read_required_payload_text(payload, "roleId", "sync_role_assets requires roleId")?;
let portrait_path = read_required_payload_text(payload, "portraitPath", "sync_role_assets requires portraitPath")?;
let generated_visual_asset_id = read_required_payload_text(payload, "generatedVisualAssetId", "sync_role_assets requires generatedVisualAssetId")?;
let generated_animation_set_id = payload.get("generatedAnimationSetId").and_then(JsonValue::as_str).map(str::trim).filter(|value| !value.is_empty()).map(ToOwned::to_owned);
let animation_map = payload.get("animationMap").cloned();
let updated_role = apply_role_asset_publish_result(&mut draft_profile, &role_id, &portrait_path, &generated_visual_asset_id, generated_animation_set_id.as_deref(), animation_map)?;
let role_name = read_optional_text_field(&updated_role, &["name"]).unwrap_or_else(|| "当前角色".to_string());
let asset_status = resolve_role_asset_status(&updated_role);
let asset_status_label = resolve_role_asset_status_label(asset_status).to_string();
upsert_asset_role_card(ctx, &session.session_id, &role_id, &updated_role, asset_status, &asset_status_label, input.submitted_at_micros)?;
let gate = summarize_publish_gate_from_json(&input.session_id, RpgAgentStage::VisualRefining, Some(&draft_profile), &parse_json_array_or_empty(&session.quality_findings_json));
let next_session = rebuild_custom_world_agent_session_row(
session,
CustomWorldAgentSessionPatch {
stage: Some(RpgAgentStage::VisualRefining),
focus_card_id: Some(Some(role_id.clone())),
draft_profile_json: Some(Some(serialize_json_value(&JsonValue::Object(draft_profile.clone()))?)),
last_assistant_reply: Some(Some(format!("已把「{}」的角色资产写回草稿,当前状态:{}", role_name, asset_status_label))),
publish_gate_json: Some(Some(serialize_json_value(&publish_gate_to_json_value(&gate))?)),
result_preview_json: Some(build_result_preview_json(Some(&draft_profile), &gate, &parse_json_array_or_empty(&session.quality_findings_json), input.submitted_at_micros)?),
checkpoints_json: Some(append_checkpoint_json(&session.checkpoints_json, &build_session_checkpoint_value("sync-role-assets", &format!("同步角色资产 {}", role_name), session))?),
asset_coverage_json: Some(build_asset_coverage_json(&draft_profile)?),
updated_at_micros: Some(input.submitted_at_micros),
..CustomWorldAgentSessionPatch::default()
},
)?;
replace_custom_world_agent_session(ctx, session, next_session);
append_custom_world_action_result_message(ctx, &session.session_id, &input.operation_id, &format!("已把「{}」的角色资产写回草稿,当前状态:{}", role_name, asset_status_label), input.submitted_at_micros);
let operation = build_and_insert_custom_world_operation(ctx, &input.operation_id, &session.session_id, RpgAgentOperationType::SyncRoleAssets, "角色资产已同步", &format!("{}」的资产状态已更新为{}", role_name, asset_status_label), input.submitted_at_micros);
Ok(build_custom_world_agent_operation_snapshot(&operation))
}
fn execute_sync_scene_assets_action(
ctx: &ReducerContext,
session: &CustomWorldAgentSession,
input: &CustomWorldAgentActionExecuteInput,
payload: &JsonMap<String, JsonValue>,
) -> Result<CustomWorldAgentOperationSnapshot, String> {
ensure_draft_refining_stage(session.stage, "sync_scene_assets")?;
let mut draft_profile = parse_optional_session_object(session.draft_profile_json.as_deref())
.ok_or_else(|| "sync_scene_assets requires an existing draft foundation".to_string())?;
let scene_id = read_required_payload_text(payload, "sceneId", "sync_scene_assets requires sceneId")?;
let scene_kind = read_required_payload_text(payload, "sceneKind", "sync_scene_assets requires sceneKind")?;
let image_src = read_required_payload_text(payload, "imageSrc", "sync_scene_assets requires imageSrc")?;
let generated_scene_asset_id = read_required_payload_text(payload, "generatedSceneAssetId", "sync_scene_assets requires generatedSceneAssetId")?;
let updated_scene = apply_scene_asset_publish_result(&mut draft_profile, &scene_id, &scene_kind, &image_src, &generated_scene_asset_id, payload.get("generatedScenePrompt").cloned().unwrap_or(JsonValue::Null), payload.get("generatedSceneModel").cloned().unwrap_or(JsonValue::Null))?;
let scene_name = read_optional_text_field(&updated_scene, &["name"]).unwrap_or_else(|| if scene_kind == "camp" { "开局营地" } else { "当前场景" }.to_string());
upsert_asset_scene_card(ctx, &session.session_id, &scene_id, &scene_kind, &updated_scene, input.submitted_at_micros)?;
let gate = summarize_publish_gate_from_json(&input.session_id, RpgAgentStage::VisualRefining, Some(&draft_profile), &parse_json_array_or_empty(&session.quality_findings_json));
let next_session = rebuild_custom_world_agent_session_row(
session,
CustomWorldAgentSessionPatch {
stage: Some(RpgAgentStage::VisualRefining),
focus_card_id: Some(Some(scene_id.clone())),
draft_profile_json: Some(Some(serialize_json_value(&JsonValue::Object(draft_profile.clone()))?)),
last_assistant_reply: Some(Some(format!("已把「{}」的场景图写回草稿,并同步刷新地点卡与幕背景状态。", scene_name))),
publish_gate_json: Some(Some(serialize_json_value(&publish_gate_to_json_value(&gate))?)),
result_preview_json: Some(build_result_preview_json(Some(&draft_profile), &gate, &parse_json_array_or_empty(&session.quality_findings_json), input.submitted_at_micros)?),
checkpoints_json: Some(append_checkpoint_json(&session.checkpoints_json, &build_session_checkpoint_value("sync-scene-assets", &format!("同步场景资产 {}", scene_name), session))?),
asset_coverage_json: Some(build_asset_coverage_json(&draft_profile)?),
updated_at_micros: Some(input.submitted_at_micros),
..CustomWorldAgentSessionPatch::default()
},
)?;
replace_custom_world_agent_session(ctx, session, next_session);
append_custom_world_action_result_message(ctx, &session.session_id, &input.operation_id, &format!("已把「{}」的场景图写回草稿,并同步刷新地点卡与幕背景状态。", scene_name), input.submitted_at_micros);
let operation = build_and_insert_custom_world_operation(ctx, &input.operation_id, &session.session_id, RpgAgentOperationType::SyncSceneAssets, "场景资产已同步", &format!("{}」的场景图已经进入当前草稿。", scene_name), input.submitted_at_micros);
Ok(build_custom_world_agent_operation_snapshot(&operation))
}
fn execute_placeholder_custom_world_action(
ctx: &ReducerContext,
session: &CustomWorldAgentSession,
@@ -3278,6 +3420,121 @@ fn parse_json_array_or_empty(raw: &str) -> Vec<JsonValue> {
.unwrap_or_default()
}
fn read_first_payload_text(payload: &JsonMap<String, JsonValue>, array_key: &str, scalar_key: &str) -> Option<String> {
payload.get(array_key).and_then(JsonValue::as_array).and_then(|values| values.first()).and_then(JsonValue::as_str)
.or_else(|| payload.get(scalar_key).and_then(JsonValue::as_str))
.map(str::trim).filter(|value| !value.is_empty()).map(ToOwned::to_owned)
}
fn find_profile_entity_by_id<'a>(profile: &'a JsonMap<String, JsonValue>, fields: &[&str], entity_id: &str) -> Option<&'a JsonMap<String, JsonValue>> {
for field in fields {
if let Some(entries) = profile.get(*field).and_then(JsonValue::as_array) {
for entry in entries {
let Some(object) = entry.as_object() else { continue; };
if read_optional_text_field(object, &["id"]).as_deref() == Some(entity_id) { return Some(object); }
}
}
}
None
}
fn apply_role_asset_publish_result(profile: &mut JsonMap<String, JsonValue>, role_id: &str, portrait_path: &str, generated_visual_asset_id: &str, generated_animation_set_id: Option<&str>, animation_map: Option<JsonValue>) -> Result<JsonMap<String, JsonValue>, String> {
for field in ["playableNpcs", "storyNpcs"] {
let Some(entries) = profile.get_mut(field).and_then(JsonValue::as_array_mut) else { continue; };
for entry in entries {
let Some(object) = entry.as_object_mut() else { continue; };
if read_optional_text_field(object, &["id"]).as_deref() != Some(role_id) { continue; }
object.insert("imageSrc".to_string(), JsonValue::String(portrait_path.to_string()));
object.insert("generatedVisualAssetId".to_string(), JsonValue::String(generated_visual_asset_id.to_string()));
if let Some(asset_id) = generated_animation_set_id { object.insert("generatedAnimationSetId".to_string(), JsonValue::String(asset_id.to_string())); }
if let Some(map) = animation_map { object.insert("animationMap".to_string(), map); }
return Ok(object.clone());
}
}
Err("目标角色不存在,无法同步角色资产。".to_string())
}
fn apply_scene_asset_publish_result(profile: &mut JsonMap<String, JsonValue>, scene_id: &str, scene_kind: &str, image_src: &str, generated_scene_asset_id: &str, generated_scene_prompt: JsonValue, generated_scene_model: JsonValue) -> Result<JsonMap<String, JsonValue>, String> {
let updated_scene = if scene_kind == "camp" {
let camp = profile.get_mut("camp").and_then(JsonValue::as_object_mut).ok_or_else(|| "目标营地不存在,无法同步场景资产。".to_string())?;
if read_optional_text_field(camp, &["id"]).as_deref() != Some(scene_id) { return Err("目标营地不存在,无法同步场景资产。".to_string()); }
camp.insert("imageSrc".to_string(), JsonValue::String(image_src.to_string()));
camp.insert("generatedSceneAssetId".to_string(), JsonValue::String(generated_scene_asset_id.to_string()));
camp.insert("generatedScenePrompt".to_string(), generated_scene_prompt);
camp.insert("generatedSceneModel".to_string(), generated_scene_model);
camp.clone()
} else {
let landmarks = profile.get_mut("landmarks").and_then(JsonValue::as_array_mut).ok_or_else(|| "目标地点不存在,无法同步场景资产。".to_string())?;
let mut updated = None;
for entry in landmarks {
let Some(object) = entry.as_object_mut() else { continue; };
if read_optional_text_field(object, &["id"]).as_deref() != Some(scene_id) { continue; }
object.insert("imageSrc".to_string(), JsonValue::String(image_src.to_string()));
object.insert("generatedSceneAssetId".to_string(), JsonValue::String(generated_scene_asset_id.to_string()));
object.insert("generatedScenePrompt".to_string(), generated_scene_prompt.clone());
object.insert("generatedSceneModel".to_string(), generated_scene_model.clone());
updated = Some(object.clone());
break;
}
updated.ok_or_else(|| "目标地点不存在,无法同步场景资产。".to_string())?
};
update_scene_chapter_acts_for_scene(profile, scene_id, image_src, generated_scene_asset_id);
Ok(updated_scene)
}
fn update_scene_chapter_acts_for_scene(profile: &mut JsonMap<String, JsonValue>, scene_id: &str, image_src: &str, generated_scene_asset_id: &str) {
let Some(chapters) = profile.get_mut("sceneChapters").and_then(JsonValue::as_array_mut) else { return; };
for chapter in chapters {
let Some(chapter_object) = chapter.as_object_mut() else { continue; };
if read_optional_text_field(chapter_object, &["sceneId"]).as_deref() != Some(scene_id) { continue; }
let Some(acts) = chapter_object.get_mut("acts").and_then(JsonValue::as_array_mut) else { continue; };
for act in acts {
if let Some(act_object) = act.as_object_mut() {
act_object.insert("backgroundImageSrc".to_string(), JsonValue::String(image_src.to_string()));
act_object.insert("backgroundAssetId".to_string(), JsonValue::String(generated_scene_asset_id.to_string()));
}
}
}
}
fn resolve_role_asset_status(role: &JsonMap<String, JsonValue>) -> CustomWorldRoleAssetStatus {
let has_portrait = read_optional_text_field(role, &["imageSrc"]).is_some() && read_optional_text_field(role, &["generatedVisualAssetId"]).is_some();
if !has_portrait { return CustomWorldRoleAssetStatus::Missing; }
let has_animation_set = read_optional_text_field(role, &["generatedAnimationSetId"]).is_some();
let has_animation_map = role.get("animationMap").and_then(JsonValue::as_object).map(|map| !map.is_empty()).unwrap_or(false);
if has_animation_set && has_animation_map { CustomWorldRoleAssetStatus::Complete } else if has_animation_set { CustomWorldRoleAssetStatus::AnimationsReady } else { CustomWorldRoleAssetStatus::VisualReady }
}
fn resolve_role_asset_status_label(status: CustomWorldRoleAssetStatus) -> &'static str {
match status { CustomWorldRoleAssetStatus::Complete => "动作已就绪", CustomWorldRoleAssetStatus::AnimationsReady => "动作补齐中", CustomWorldRoleAssetStatus::VisualReady => "主图已就绪", CustomWorldRoleAssetStatus::Missing => "待生成主图" }
}
fn build_asset_coverage_json(profile: &JsonMap<String, JsonValue>) -> Result<String, String> {
serialize_json_value(&json!({"roleAssets": [], "sceneAssets": [], "allRoleAssetsReady": false, "allSceneAssetsReady": false, "profileId": read_optional_text_field(profile, &["id"])}))
}
fn upsert_asset_role_card(ctx: &ReducerContext, session_id: &str, role_id: &str, role: &JsonMap<String, JsonValue>, asset_status: CustomWorldRoleAssetStatus, asset_status_label: &str, updated_at_micros: i64) -> Result<(), String> {
let title = read_optional_text_field(role, &["name"]).unwrap_or_else(|| "角色".to_string());
let subtitle = read_optional_text_field(role, &["role", "relationToPlayer", "publicMask"]).unwrap_or_else(|| asset_status_label.to_string());
let summary = read_optional_text_field(role, &["summary", "description", "publicMask"]).unwrap_or_else(|| "角色资产已写回草稿。".to_string());
upsert_asset_card(ctx, session_id, role_id, RpgAgentDraftCardKind::Character, &title, &subtitle, &summary, Some(asset_status), Some(asset_status_label), updated_at_micros)
}
fn upsert_asset_scene_card(ctx: &ReducerContext, session_id: &str, scene_id: &str, scene_kind: &str, scene: &JsonMap<String, JsonValue>, updated_at_micros: i64) -> Result<(), String> {
let kind = if scene_kind == "camp" { RpgAgentDraftCardKind::Camp } else { RpgAgentDraftCardKind::Landmark };
let title = read_optional_text_field(scene, &["name"]).unwrap_or_else(|| if scene_kind == "camp" { "开局营地" } else { "场景" }.to_string());
let subtitle = read_optional_text_field(scene, &["purpose", "mood", "dangerLevel"]).unwrap_or_else(|| "场景资产已就绪".to_string());
let summary = read_optional_text_field(scene, &["summary", "description", "publicMask"]).unwrap_or_else(|| "场景图已写回草稿。".to_string());
upsert_asset_card(ctx, session_id, scene_id, kind, &title, &subtitle, &summary, None, Some("场景图已就绪"), updated_at_micros)
}
fn upsert_asset_card(ctx: &ReducerContext, session_id: &str, card_id: &str, kind: RpgAgentDraftCardKind, title: &str, subtitle: &str, summary: &str, asset_status: Option<CustomWorldRoleAssetStatus>, asset_status_label: Option<&str>, updated_at_micros: i64) -> Result<(), String> {
let detail_payload = json!({"id": card_id, "kind": kind.as_str(), "title": title, "sections": [{"id": "summary", "label": "摘要", "value": summary}], "linkedIds": [], "locked": false, "editable": true, "editableSectionIds": ["summary"], "warningMessages": []});
let next = CustomWorldDraftCard { card_id: card_id.to_string(), session_id: session_id.to_string(), kind, status: RpgAgentDraftCardStatus::Draft, title: title.to_string(), subtitle: subtitle.to_string(), summary: summary.to_string(), linked_ids_json: "[]".to_string(), warning_count: 0, asset_status, asset_status_label: asset_status_label.map(ToOwned::to_owned), detail_payload_json: Some(serialize_json_value(&detail_payload)?), created_at: Timestamp::from_micros_since_unix_epoch(updated_at_micros), updated_at: Timestamp::from_micros_since_unix_epoch(updated_at_micros) };
if let Some(existing) = ctx.db.custom_world_draft_card().card_id().find(&card_id.to_string()).filter(|entry| entry.session_id == session_id) { replace_custom_world_draft_card(ctx, &existing, CustomWorldDraftCard { created_at: existing.created_at, ..next }); } else { ctx.db.custom_world_draft_card().insert(next); }
Ok(())
}
fn serialize_json_value(value: &JsonValue) -> Result<String, String> {
serde_json::to_string(value).map_err(|error| format!("JSON 序列化失败: {error}"))
}

View File

@@ -1192,6 +1192,25 @@ pub fn get_custom_world_agent_operation(
}
}
#[spacetimedb::procedure]
pub fn upsert_custom_world_agent_operation_progress(
ctx: &mut ProcedureContext,
input: CustomWorldAgentOperationProgressInput,
) -> CustomWorldAgentOperationProcedureResult {
match ctx.try_with_tx(|tx| upsert_custom_world_agent_operation_progress_tx(tx, input.clone())) {
Ok(operation) => CustomWorldAgentOperationProcedureResult {
ok: true,
operation: Some(operation),
error_message: None,
},
Err(message) => CustomWorldAgentOperationProcedureResult {
ok: false,
operation: None,
error_message: Some(message),
},
}
}
fn continue_story_tx(
ctx: &ReducerContext,
input: StoryContinueInput,
@@ -1474,6 +1493,59 @@ fn get_custom_world_agent_operation_tx(
Ok(build_custom_world_agent_operation_snapshot(&operation))
}
fn upsert_custom_world_agent_operation_progress_tx(
ctx: &ReducerContext,
input: CustomWorldAgentOperationProgressInput,
) -> Result<CustomWorldAgentOperationSnapshot, String> {
validate_custom_world_agent_operation_progress_input(&input).map_err(|error| error.to_string())?;
ctx.db
.custom_world_agent_session()
.session_id()
.find(&input.session_id)
.filter(|row| row.owner_user_id == input.owner_user_id)
.ok_or_else(|| "custom_world_agent_session 不存在".to_string())?;
let timestamp = Timestamp::from_micros_since_unix_epoch(input.updated_at_micros);
let operation = if let Some(current) = ctx
.db
.custom_world_agent_operation()
.operation_id()
.find(&input.operation_id)
{
if current.session_id != input.session_id {
return Err("custom_world_agent_operation.session_id 不匹配".to_string());
}
let next = rebuild_custom_world_agent_operation_row(
&current,
CustomWorldAgentOperationPatch {
status: Some(input.operation_status),
phase_label: Some(input.phase_label.clone()),
phase_detail: Some(input.phase_detail.clone()),
progress: Some(input.operation_progress),
error_message: Some(input.error_message.clone()),
updated_at_micros: Some(input.updated_at_micros),
},
)?;
replace_custom_world_agent_operation(ctx, &current, next.clone());
next
} else {
ctx.db.custom_world_agent_operation().insert(CustomWorldAgentOperation {
operation_id: input.operation_id.clone(),
session_id: input.session_id.clone(),
operation_type: input.operation_type,
status: input.operation_status,
phase_label: input.phase_label.clone(),
phase_detail: input.phase_detail.clone(),
progress: input.operation_progress,
error_message: input.error_message.clone(),
created_at: timestamp,
updated_at: timestamp,
})
};
Ok(build_custom_world_agent_operation_snapshot(&operation))
}
fn finalize_custom_world_agent_message_turn_tx(
ctx: &ReducerContext,
input: CustomWorldAgentMessageFinalizeInput,
@@ -2896,14 +2968,22 @@ fn execute_custom_world_agent_action_tx(
.filter(|row| row.owner_user_id == input.owner_user_id)
.ok_or_else(|| "custom_world_agent_session 不存在".to_string())?;
if ctx
if let Some(existing_operation) = ctx
.db
.custom_world_agent_operation()
.operation_id()
.find(&input.operation_id)
.is_some()
{
return Err("custom_world_agent_operation.operation_id 已存在".to_string());
let can_reuse_running_draft_operation = input.action.trim() == "draft_foundation"
&& existing_operation.session_id == input.session_id
&& existing_operation.operation_type == RpgAgentOperationType::DraftFoundation
&& matches!(
existing_operation.status,
RpgAgentOperationStatus::Queued | RpgAgentOperationStatus::Running
);
if !can_reuse_running_draft_operation {
return Err("custom_world_agent_operation.operation_id 已存在".to_string());
}
}
let payload = parse_optional_session_object(input.payload_json.as_deref()).unwrap_or_default();
@@ -2990,7 +3070,7 @@ fn execute_draft_foundation_action(
updated_at,
);
let operation = build_and_insert_custom_world_operation(
let operation = complete_custom_world_operation(
ctx,
&input.operation_id,
&session.session_id,
@@ -2998,7 +3078,7 @@ fn execute_draft_foundation_action(
"底稿已整理",
"第一版 foundation draft 已写入会话与世界卡。",
updated_at,
);
)?;
Ok(build_custom_world_agent_operation_snapshot(&operation))
}
@@ -4117,6 +4197,53 @@ fn replace_custom_world_draft_card(
ctx.db.custom_world_draft_card().insert(next);
}
fn complete_custom_world_operation(
ctx: &ReducerContext,
operation_id: &str,
session_id: &str,
operation_type: RpgAgentOperationType,
phase_label: &str,
phase_detail: &str,
timestamp_micros: i64,
) -> Result<CustomWorldAgentOperation, String> {
if let Some(current) = ctx
.db
.custom_world_agent_operation()
.operation_id()
.find(&operation_id.to_string())
{
if current.session_id != session_id {
return Err("custom_world_agent_operation.session_id 不匹配".to_string());
}
if current.operation_type != operation_type {
return Err("custom_world_agent_operation.operation_type 不匹配".to_string());
}
let next = rebuild_custom_world_agent_operation_row(
&current,
CustomWorldAgentOperationPatch {
status: Some(RpgAgentOperationStatus::Completed),
phase_label: Some(phase_label.to_string()),
phase_detail: Some(phase_detail.to_string()),
progress: Some(100),
error_message: Some(None),
updated_at_micros: Some(timestamp_micros),
},
)?;
replace_custom_world_agent_operation(ctx, &current, next.clone());
return Ok(next);
}
Ok(build_and_insert_custom_world_operation(
ctx,
operation_id,
session_id,
operation_type,
phase_label,
phase_detail,
timestamp_micros,
))
}
fn build_and_insert_custom_world_operation(
ctx: &ReducerContext,
operation_id: &str,

View File

@@ -112,3 +112,20 @@ test('creation hub mixes puzzle works into the same grid and uses puzzle tag to
expect(screen.getAllByText('拼图').length).toBeGreaterThan(0);
expect(screen.queryByText('我的拼图作品')).toBeNull();
});
test('creation hub shows delete action for persisted rpg drafts', () => {
render(
<CustomWorldCreationHub
items={[{ ...baseDraftItem, profileId: 'profile-1' }]}
loading={false}
error={null}
onRetry={() => {}}
onCreateType={noopCreateType}
onOpenDraft={() => {}}
onEnterPublished={() => {}}
onDeletePublished={() => {}}
/>,
);
expect(screen.getByRole('button', { name: '删除' })).toBeTruthy();
});

View File

@@ -1,4 +1,4 @@
import { useMemo, useState } from 'react';
import { useMemo, useState } from 'react';
import type { CustomWorldWorkSummary } from '../../../packages/shared/src/contracts/customWorldAgent';
import type { BigFishWorkSummary } from '../../../packages/shared/src/contracts/bigFishWorkSummary';
@@ -199,9 +199,7 @@ export function CustomWorldCreationHub({
: null
}
onDelete={
item.kind === 'rpg' &&
item.item.status === 'published' &&
item.item.profileId
item.kind === 'rpg' && item.item.profileId
? () => {
onDeletePublished?.(item.item);
}

View File

@@ -1216,6 +1216,42 @@ export function PlatformEntryFlowShellImpl({
],
);
const handleDeleteLibraryEntry = useCallback(
(entry: CustomWorldLibraryEntry<CustomWorldProfile>) => {
if (!entry.profileId || deletingCreationWorkId) {
return;
}
runProtectedAction(() => {
const confirmed = window.confirm(
`确认删除作品《${entry.worldName}》吗?删除后会从你的作品列表和公开广场中移除。`,
);
if (!confirmed) {
return;
}
setDeletingCreationWorkId(entry.profileId);
platformBootstrap.setPlatformError(null);
void deleteRpgEntryWorldProfile(entry.profileId)
.then(async (entries) => {
platformBootstrap.setSavedCustomWorldEntries(entries);
await platformBootstrap.refreshCustomWorldWorks().catch(() => []);
await platformBootstrap.refreshPublishedGallery().catch(() => []);
})
.catch((error) => {
platformBootstrap.setPlatformError(
resolveRpgCreationErrorMessage(error, '删除自定义世界失败。'),
);
})
.finally(() => {
setDeletingCreationWorkId(null);
});
});
},
[deletingCreationWorkId, platformBootstrap, runProtectedAction],
);
const handleDeletePublishedWork = useCallback(
(work: (typeof creationHubItems)[number]) => {
if (!work.profileId || deletingCreationWorkId) {
@@ -1556,6 +1592,10 @@ export function PlatformEntryFlowShellImpl({
detailNavigation.openLibraryDetail(entry);
});
}}
onDeleteLibraryEntry={(entry) => {
handleDeleteLibraryEntry(entry);
}}
deletingLibraryEntryId={deletingCreationWorkId}
onSearchPublicCode={(keyword) => {
void handlePublicCodeSearch(keyword);
}}

View File

@@ -76,6 +76,10 @@ export interface RpgEntryHomeViewProps {
onOpenLibraryDetail: (
entry: CustomWorldLibraryEntry<CustomWorldProfile>,
) => void;
onDeleteLibraryEntry?: (
entry: CustomWorldLibraryEntry<CustomWorldProfile>,
) => void;
deletingLibraryEntryId?: string | null;
onSearchPublicCode?: (keyword: string) => void | Promise<void>;
isSearchingPublicCode?: boolean;
onOpenProfileDashboardCard?: (cardKey: ProfileDashboardCardKey) => void;
@@ -303,9 +307,13 @@ function WorldCard({
function CreationLibraryCard({
entry,
onClick,
onDelete,
isDeleting = false,
}: {
entry: CustomWorldLibraryEntry<CustomWorldProfile>;
onClick: () => void;
onDelete?: () => void;
isDeleting?: boolean;
}) {
const coverImage = resolvePlatformWorldCoverImage(entry);
const leadPortrait = resolvePlatformWorldLeadPortrait(entry);
@@ -343,6 +351,19 @@ function CreationLibraryCard({
/>
) : null}
<div className="absolute inset-0 bg-[var(--platform-card-overlay-strong)]" />
{onDelete ? (
<button
type="button"
onClick={(event) => {
event.stopPropagation();
onDelete();
}}
disabled={isDeleting}
className="platform-button platform-button--danger absolute right-2 top-2 z-20 min-h-0 rounded-full px-2.5 py-1 text-[10px] disabled:cursor-not-allowed disabled:opacity-60"
>
{isDeleting ? '删除中' : '删除'}
</button>
) : null}
<div className="relative z-10 flex h-full min-w-0 flex-col">
<div className="flex min-w-0 flex-wrap items-center gap-1.5">
<span
@@ -784,6 +805,8 @@ export function RpgEntryHomeView({
onOpenCreateTypePicker,
onOpenGalleryDetail,
onOpenLibraryDetail,
onDeleteLibraryEntry,
deletingLibraryEntryId = null,
onSearchPublicCode,
isSearchingPublicCode = false,
onOpenProfileDashboardCard,
@@ -983,6 +1006,8 @@ export function RpgEntryHomeView({
key={`${entry.ownerUserId}:${entry.profileId}:mine`}
entry={entry}
onClick={() => onOpenLibraryDetail(entry)}
onDelete={onDeleteLibraryEntry ? () => onDeleteLibraryEntry(entry) : undefined}
isDeleting={deletingLibraryEntryId === entry.profileId}
/>
),
)}