zizi-al/games-development-ai
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
894 Commits 4 Branches 0 Tags
Commit Graph

894 Commits

This Branch
This Branch
All Branches
Author SHA1 Message Date
lili
c5ee8ef5e9 feat(cheap): 便宜档 Service 收窄工具面——关 Planning/Team/Schedule 三组框架工具(工单 h) 
2.0.2 get_toolkit(app/_service/_toolkit.py)除 workspace 内建外,还无条件并入 Planning
(Task* :109-110 "always on")/Team(Team*·AgentCreate :144-168),并按 session model 挂
Schedule(schedule_tools 组 :117-142)。便宜档单机单游戏用不到,纯工具面噪声(80011 flail:
12+ 工具语义重叠是根因之一;护城河 I1 六工具白名单邻域治理)。

修法:cheap Service 进程内 monkeypatch 包裹 _chat.get_toolkit(_chat.py:33 本地绑定、:240 唯一
装配调用者),调原函数拿到 Toolkit 后重建干净版:basic 组按 .name 剔 Planning+Team、丢整个
schedule_tools 组;workspace skills/MCP、六工具、后台控制 ToolStop 一律原样保留。钉 agentscope
==2.0.2、版本漂移响亮告警仍应用;作用域=cheap Service 进程(只跑便宜档 agent),tier2 独立进程
不受影响;绝不改 venv/site-packages 本体。惰性 import _chat(牵 apscheduler)保无 [service] 环境
可 import 本模块。

测试(tests/test_planning_team_tools_patch.py):4 filter/wrapper 单测(补丁后无三组 / 保六工具+
ToolStop / 过滤幂等 / 装配幂等)本机全绿 + 1 真 _chat 集成测 importorskip apscheduler(本机跳、
mini-desktop 全跑)。全套 pytest tests/ -q = 333 passed, 1 skipped。

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
 2026-07-04 00:46:09 -07:00
lili
bfeaea2d42 feat(tier2): 创始人令提生成预算阈值——writer 40→60 轮 / step_cap 60→100 / model_calls 80→120 / 墙钟 1800→2700s
Some checks failed
contract-gates / contract-gates (push) Has been cancelled
Details
docs-gate / docs-gate (push) Has been cancelled
Details 
F-2 两轮实证 40/60 是预算墙非能力墙:美食类硬 brief R1 churn 到 step_cap、
R2 设计收敛后 writer 仍 40 轮被截,resume 续修窗只剩 ~20 步修不完。步顶 100
= writer 60 + 续修 ~40 headroom;失控保护交 ¥ 两段式(50/75)与墙钟 2700。
cheap Service 显式传 150 不受影响;cheap CLI 继承(纯兜底,¥10/15 是真闸)。
tier2 97/97 + cheap 329/329。

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
 2026-07-04 00:14:32 -07:00
lili
0a9504e400 fix(tier2): 设计团队墙钟 240→420s(F-2 R1 两败局实证,KTD2 定点修)
Some checks failed
contract-gates / contract-gates (push) Has been cancelled
Details
docs-gate / docs-gate (push) Has been cancelled
Details 
R1=3/5:两败局(美食/糖水店)同构——3/4 专家已产出、末段调用/leader 撞 240s
墙钟→降级单 agent 设计→writer 40 轮不收敛→step_cap;三过门局设计团队全部
正常收敛(超时↔失败 5/5 完美相关,台账 fail_system 空导致退路树误判分散)。
M3 thinking 慢尾单调用可达 120-180s,420 给一次慢尾余量,与 step_timeout_s
同刻度。YAML(运行面)与 _BUILTIN_DEFAULTS(兜底)同改防漂移;tier2 97/97。

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
 2026-07-03 23:56:32 -07:00
lili
e01d91fb83 feat(quality): 经营品类金标正例固化入仓 bake-shop-serve-1(欠账①清)+ 骨架转薄对照定案(欠账②按红线收束)
Some checks failed
contract-gates / contract-gates (push) Has been cancelled
Details
docs-gate / docs-gate (push) Has been cancelled
Details 
- 正例=S5b bake_off 富化过门款(烘焙上菜 rush,attempts=1/¥2.00/tap-targets 真玩过门):src 7 文件+run-summary+richness-n3 凭据入 fixtures/golden-samples/sim-business/
- 12 条口径 n=3 复采样中位 10/12(单发 11/9/10 方差±2 合纪律),对照骨架 4/11 通用段分离 5–6 分=通用尺判别锚
- 品类段诚实口径:2/4 vs 1/4 仅 1 分分离(serve 形态无进货/双币),品类段正锚待进货型经营款,回退判定以通用段+L4 为准
- 合成薄壳反例不再补:骨架实测即真实薄对照,不臆造材料充数
- rubric _note 收束两欠账 + skill 金标锚点段 + golden-samples README 段;rubric-sync-gate 过

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
 2026-07-03 23:07:26 -07:00
lili
b0bfcabf8a docs(cutover): §9 S6 负例四门真机证据(日额/背压/暂停+并发)+ 六门放行证据表(prod/lab 口径明标)+ 工单回写 a1/a2/e
Some checks failed
contract-gates / contract-gates (push) Has been cancelled
Details
docs-gate / docs-gate (push) Has been cancelled
Details 
- D12 五种拒绝行为全负例双证(HTTP 码+后端日志):并发/日额(当日已用=7 上限=0)/背压(门②全局在飞=0)/暂停(infra_config 热键)/unsafe
- 六门证据表集齐:9d trace/D11(86/90)/D9/firstPlay/GP9/D12 全 prod 口径;S5b 明标 lab;放行+S5a=创始人
- 工单:a1 ✅(C6 族错配直指 f121bb01)/ a2 ✅关闭(前提证伪:模板 targets 本就齐,是生成丢键)/ e ✅(2a8d5e45)

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
 2026-07-03 22:58:52 -07:00
lili
48739533e5 docs(plans): cutover §9 S5b 落分——lab 复验达标(三品类收敛局 12/12 全过,raw 80% 亦过线;修复红利成本 ¥0.7-3.7);S5a 生产分布回填留创始人 (cutover 窗口)
Some checks failed
contract-gates / contract-gates (push) Has been cancelled
Details
docs-gate / docs-gate (push) Has been cancelled
Details 
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
 2026-07-03 22:32:46 -07:00
lili
5e4d81a296 merge: C1金标欠账清偿——narrative 12条正式重锚+trpg/puzzle富正例基线(前提证伪:模板=黄金脚手架非薄反)+判别主锚=品类段+L4 归质量轨 (尾波C1欠账)
Some checks failed
contract-gates / contract-gates (push) Has been cancelled
Details
docs-gate / docs-gate (push) Has been cancelled
Details 
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
 2026-07-03 21:34:36 -07:00
lili
b34a7725cc docs(.agents): C1金标欠账清偿——narrative 12条口径正式重锚(正8/12 vs 薄反7/12,判别主锚=品类段+L4)+ trpg/puzzle 实为黄金脚手架富正例、基线入仓(真薄反欠账明记不现造,规范三合成薄壳教训在案)+ fixture _note 同步收口(复采样×3取中位) (尾波C1欠账) 
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
 2026-07-03 21:34:22 -07:00
lili
42eed68675 merge: test_a11 m3/m4 模块级 game_dir/dedup 替换收进可恢复 fixture——根除 5 个跨模块污染失败(证据修正:基线12≠全污染,余7=环境前置) (尾波小单)
Some checks failed
contract-gates / contract-gates (push) Has been cancelled
Details
docs-gate / docs-gate (push) Has been cancelled
Details 
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
 2026-07-03 21:13:06 -07:00
lili
11436f07cc test(cheap): test_a11_m3/m4 模块级 game_dir/dedup 替换收进可恢复 autouse fixture——根除 game_dir 跨模块隔离污染(基线二分钉死:5 个失败,非 12) (尾波小单) 
test_a11_m3_deterministic 与 test_a11_m4_regenerate 在 import 时无恢复地替换
cheap_run.game_dir(两文件同构),污染同会话后续测试模块:受害者按真 games/ 根校验
write 边界,game_dir 被改指临时目录后一写即判越界。改为 module 级 autouse fixture
(yield 后恢复原值),__main__ 独立运行路径同装(单进程无需恢复),dedup 登记表一并收拢。

证据:全套 12→7 failed、三种文件顺序结果恒定(7 failed / 322 passed);二分钉死纯
game_dir 污染 = 5 个,均已恢复 = test_run::test_scaffold + test_toolkit::{write_l1_rejected,
write_l3_ok,check_shape_drawbutton_ok,check_shape_define_fail}。

余 7 个非本污染、隔离下同红,系环境前置:worktree 无 esbuild →
test_run::{test_stage,test_play_real}、test_toolkit::test_check_and_build_clean_pass;
缺真实 e2e 产物 → test_studio_smoke×4(须先跑 cheap_studio 单局)。须集成段 mini-desktop
才具备,未纳入本单。同族但未致失败的 test_worker_service 模块级 dedup 泄漏按最小改动未动。

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
 2026-07-03 21:11:31 -07:00
lili
e524adfb0d docs(.agents): wave-close 第8步——docs-gate 六检改七检(G7)+merge commit 手跑 B8 条件门清单(pre-commit 不触发缺口) (尾波挂点)
Some checks failed
contract-gates / contract-gates (push) Has been cancelled
Details
docs-gate / docs-gate (push) Has been cancelled
Details 
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
 2026-07-03 21:02:00 -07:00
lili
4ec86b2524 docs(.agents): trpg §10 通用条数升 12 条口径(L2×7+第12条入列)+ narrative 金标基线标注旧尺实测与重锚欠账(史实不改) (尾波措辞收敛)
Some checks failed
contract-gates / contract-gates (push) Has been cancelled
Details
docs-gate / docs-gate (push) Has been cancelled
Details 
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
 2026-07-03 20:58:57 -07:00
lili
7820c75b8b docs(contracts): prompts README 死链修——退役执行档改 git 史指针(留痕纪律) (尾波)
Some checks failed
contract-gates / contract-gates (push) Has been cancelled
Details
docs-gate / docs-gate (push) Has been cancelled
Details 
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
 2026-07-03 20:54:46 -07:00
lili
41cde493ee merge: 尾波SVG批——plan①顶图纵切redraw(cutover三态入图)+质量SoT quality-01(12条口径redraw)/02(rubric+F-4五裁定新图)+重推演档§0门面图;4图atlas规范+XML复验+PNG眼验全过(眼验抓获A11格过时+紫盒溢出已修) (尾波)
Some checks failed
contract-gates / contract-gates (push) Has been cancelled
Details
docs-gate / docs-gate (push) Has been cancelled
Details 
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
 2026-07-03 20:51:48 -07:00
lili
e6447efc40 docs(svg): 总览图眼验修正——A11 计费 e2e 格转现行绿(07-04 已收官)+SAA 紫盒末行溢出补高(眼验抓获) (尾波SVG批) 
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
 2026-07-03 20:51:31 -07:00
lili
2b9ebdbda4 docs(svg): plan①顶图纵切redraw+质量SoT quality-01/02门面图+重推演档§0门面图(atlas规范,眼验待主会话) (尾波SVG批) 
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
 2026-07-03 20:41:50 -07:00
lili
ee48cc2de3 docs(plans): cutover §9 S5b——两次假起步各钉一真缺陷(conc争用测量污染/CHROME_BIN盲注=生产考卷回退key-cycle共同上游根因e4a4d025);第三跑在飞 (cutover 窗口)
Some checks failed
contract-gates / contract-gates (push) Has been cancelled
Details
docs-gate / docs-gate (push) Has been cancelled
Details 
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
 2026-07-03 20:31:05 -07:00
lili
e4a4d02516 fix(cheap): CHROME_BIN 只在 Mac 路径真实存在时注入——Linux 上盲注致 smoke 全体 spawn ENOENT→state=None→考卷全回退 key-cycle(80007 族错配盲修¥14.9/80012 driver=key-cycle 共同上游根因;S5b 串行复跑+raw 观测行连环钉死) (cutover S5b)
Some checks failed
contract-gates / contract-gates (push) Has been cancelled
Details
docs-gate / docs-gate (push) Has been cancelled
Details 
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
 2026-07-03 20:24:44 -07:00
lili
8a58d57972 docs(.agents): tech-decisions §13 蒸馏——框架默认注入面击穿护城河(第三例跨路径封口·内建工具面红线)+ stuck签名承载同错语义+失败即时归因 + 运维三硬记(jar世代=env面随升/快照验完整性/并发争用污染测量) (cutover 窗蒸馏)
Some checks failed
contract-gates / contract-gates (push) Has been cancelled
Details
docs-gate / docs-gate (push) Has been cancelled
Details 
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
 2026-07-03 20:21:18 -07:00
lili
396461065b chore(cheap): smoke FAIL 打印 raw 尾(bake_off 并发争用致 100% 回退考卷、真因被吞实证)——失败真因可见纪律 (cutover 窗口)
Some checks failed
contract-gates / contract-gates (push) Has been cancelled
Details
docs-gate / docs-gate (push) Has been cancelled
Details 
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
 2026-07-03 20:15:05 -07:00
lili
0ca5320988 docs(mvp): 板对账——W-S1 cutover S0-S4+回滚演练完成/W-A11 ✅ 收官/人办③④已执行(证据指针=cutover plan §9) (cutover 窗口)
Some checks failed
contract-gates / contract-gates (push) Has been cancelled
Details
docs-gate / docs-gate (push) Has been cancelled
Details 
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
 2026-07-03 20:06:38 -07:00
lili
b79625e0f5 docs(plans): cutover §9 回滚演练全绿(幂等重放终态保护/验签双向/off旁路on恢复全循环/透传修生产实证)——S6 回滚判据达成 (cutover 窗口)
Some checks failed
contract-gates / contract-gates (push) Has been cancelled
Details
docs-gate / docs-gate (push) Has been cancelled
Details 
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
 2026-07-03 20:04:56 -07:00
lili
c6d81fc287 docs(plans): cutover §9——A11 计费 e2e 收官(93150/D11=90/cost 0.9871 落账/C5 考卷重生 modify 路首证);S5b bake_off lab 复验已起跑 (cutover 窗口)
Some checks failed
contract-gates / contract-gates (push) Has been cancelled
Details
docs-gate / docs-gate (push) Has been cancelled
Details 
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
 2026-07-03 19:52:56 -07:00
lili
5cc79de871 docs(plans): cutover §9 S4 绿腿证据(80012/¥2.13/repairs=1/D11=86/engineBundle 238KB)+A11 同窗腿(jar轮换@2a8d5e45/Nacos凭据双杀教训/两段式回路生产真转) (cutover 窗口)
Some checks failed
contract-gates / contract-gates (push) Has been cancelled
Details
docs-gate / docs-gate (push) Has been cancelled
Details 
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
 2026-07-03 19:48:23 -07:00
lili
2a8d5e450b chore(cheap): 丰富度日志行补 genre 段(工单e,纯观测;genre 实参本就透传) (cutover S2)
Some checks failed
contract-gates / contract-gates (push) Has been cancelled
Details
docs-gate / docs-gate (push) Has been cancelled
Details 
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
 2026-07-03 19:24:35 -07:00
lili
f121bb01ce merge: C6反馈驱动器族错配直指(回退selectionBasis+H零增量→先修targets/spec别改计分;80007盲修¥14.9对症解)+a2前提证伪实录(shop模板targets一直在,回退发生在生成物丢键) (cutover S2 工单a) 
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
 2026-07-03 19:24:08 -07:00
lili
c814668992 docs(plans): cutover §9 续记——工具面击穿发现(内建旁路写白名单,补丁a6d5c29a)+80011兜底链组合实证(异类失败零误熔断/软停/step_cap/诚实回调) (cutover 窗口)
Some checks failed
contract-gates / contract-gates (push) Has been cancelled
Details
docs-gate / docs-gate (push) Has been cancelled
Details 
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
 2026-07-03 19:23:10 -07:00
lili
8d94128d7e fix(cheap): C6反馈驱动器族错配直指(selectionBasis回退+H零增量→先修targets/spec别改计分) (cutover S2 工单a) 
工单a1(真修):build_cheap_feedback_obj 的 H_progress 分支——当判卷驱动器族是「回退」兜出来的
(staged play-spec.json 的 driver.selectionBasis 含「回退」二字)且 H 失败为零增量类(同源读 g.checks
的 before/after 判 after<=before,不 re-parse 字符串)时,把该门 fixDirection 从原「核对得分随真玩上升」
掉转为直指:先让运行态暴露 targets / 修 play-spec,而不是改计分逻辑;原 H 详情保留在 detail。
三条不命中支(driver 正常选中 / score 有真增即终局不可达 / 读不到 selectionBasis)一律保原门提示一字不变。
新增 _cheap_selection_basis(镜像 _cheap_driver_type 容错)、_h_zero_increment 两个 helper。
80007 盲修¥14.9(key-cycle 驱不动购销循环→真玩零得分→原文案指向改计分→M3 盲修烧到硬地板)实证。
单测 test_gate_judge_mismatch.py 7 用例全绿;既有 gate_judge/feedback 32 用例零回归;tier2 路未碰。

工单a2(实证无需改):前提「_template-shop 漏 targets 契约」经端到端真跑证伪。
_template-shop 自建档提交 9ddd173c(2026-06-30)起,_forensicsView().state() 就暴露 targets 键,与参考
模板 _template-feiyi 结构一致;A2 批(34923812)只加「起局按钮解耦 render + targets 键名约定进 prompt」、
未动 targets 暴露本身。全管线 scaffold→build→stage→smoke→ensure_play_spec 真跑,shop 得
driver.type=tap-targets(selectionBasis=可点目标类,非 key-cycle 回退),与 feiyi 正对照行为一致。
模板已满足契约,按最小改动不造多余改动。

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
 2026-07-03 19:21:55 -07:00
lili
a6d5c29ad3 fix(cheap): Service 关 workspace 内建工具(Bash/Edit/Glob/Grep/Read/Write)——内建 Write/Bash 旁路六工具写白名单(护城河击穿)+ M3 12+工具面语义重叠混乱税(80009 死圈/80011 失败目录:Bash CWD 混乱/read_file 被串 Read 参数/Edit 连败,签名富化观测实证) (cutover S2 工具面封口)
Some checks failed
contract-gates / contract-gates (push) Has been cancelled
Details
docs-gate / docs-gate (push) Has been cancelled
Details 
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
 2026-07-03 19:12:35 -07:00
lili
daee094c83 docs(plans): cutover plan §9 执行记录——S0-S3 证据落档(S1 worker接线bug/S2 GP9+并发+Bean功能实证+studio透传+80009熔断观测盲区双修/S3 起步值对账+BUDGET_RMB废止)+残余工单七件 (cutover 窗口)
Some checks failed
contract-gates / contract-gates (push) Has been cancelled
Details
docs-gate / docs-gate (push) Has been cancelled
Details 
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
 2026-07-03 19:05:02 -07:00
lili
d3ec652d82 merge: studio透传控制面三拒绝码(quota/backpressure/unsafe)保留产品语义 (cutover S2 工单d)
Some checks failed
contract-gates / contract-gates (push) Has been cancelled
Details
docs-gate / docs-gate (push) Has been cancelled
Details 
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
 2026-07-03 18:53:39 -07:00
lili
ff66c1c806 merge: stuck熔断签名富化+工具失败即时观测(80009误熔断+归因盲区双修) (cutover S2) 
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
 2026-07-03 18:53:39 -07:00
lili
5afb9823cd fix(tier2): stuck 熔断签名富化=工具名+状态+归一化错误头,对齐设计意图「同错反复」——异类4连败不再误熔断(交step_cap/¥闸兜底)+工具失败即时打印观测(80009 生产事故:签名恒为裸error且无从归因) (cutover S2 观测修) 
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
 2026-07-03 18:53:09 -07:00
lili
f9b85d429b fix(studio): 透传控制面三拒绝码(1101001001/002/004)保留产品语义,其余异常仍包裹 1112002000 (cutover S2 工单d) 
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
 2026-07-03 18:51:08 -07:00
lili
dbfe40906e fix(cheap): worker_service 门面路 _service_run_fn 先导 _bootstrap 再导 driver——补 tier2/gen-worker sys.path 接线(CLI 路有、门面路漏;S1 真验实证 No module named worker 兜底 failed 回调) (cutover S1)
Some checks failed
contract-gates / contract-gates (push) Has been cancelled
Details
docs-gate / docs-gate (push) Has been cancelled
Details 
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
 2026-07-03 16:07:48 -07:00
lili
30024bd134 merge: C1 金标复验五品类(12条口径,真M3 ¥0.46)——第12条对前11条零漂移坐实(heritage精确×2/trpg精确/puzzle±1/narrative纯隔离精确复现旧锚);narrative薄反触线经复采样n=3定性=旧锚stingy非item12效应(M3单发方差实测±2~4,复采样条款首战生效);5 fixture _note 追加式新基线,items零动;欠账=narrative正式重锚(固定brief+genre口径)/trpg·puzzle薄反入仓/经营正例待固化
Some checks failed
contract-gates / contract-gates (push) Has been cancelled
Details
docs-gate / docs-gate (push) Has been cancelled
Details 
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
 2026-07-03 15:27:31 -07:00
lili
2aafa82b9c docs(cheap): 通用底座第12条金标复验回写5品类rubric _note(12条口径·C波·真M3 23调用¥0.46) 
评分尺变更「通用底座 11→12 条(⑨升「核心操作非无脑」L2·追加末尾·前11一字不动)」的金标复验:
逐品类金标正例+薄反例真 M3 judge 复评,比对前 11 条分组小计对旧 11 条锚的漂移(±1 线),
第 12 条命中态与品类段单记。结论=item12 对前 11 零漂移坐实:

- heritage 正/薄反 前11 逐值精确=旧锚(10/11=6/2/2·6/11=4/0/2),品类 5/5·3/5 均±1内
- trpg 正例 前11 总分 8=旧锚(首记分组 5/1/2),品类 5/5 持平
- puzzle 正例(_fewshot)9/11 首记通用、对照(_template)8/11 对旧锚7/11 +1(±1内),品类 5/5 持平
- narrative 正例 双口径(生产/纯隔离)中位均±1内,genre=None 纯item12隔离中位7/11精确复现旧通用锚
- 第12条判别力真实:富/决策游戏命中、经营骨架(点顾客自动结算)与故事骨架多未命中;sim_neg item12=False

四品类(heritage/trpg/puzzle)+narrative正例 12条口径新基线已回写(追加式·保留旧11条记录)。

待创始人裁 / 欠账:
1. narrative 薄反例 L2 中位 2~3 对旧锚 L2 1 超 +1~+2 —— 根因旧锚系「骨架原样」空brief单发stingy观测
   (该骨架恒有 juice+playSfx 两个L2、真实地板≥2),非 item12 效应;narrative 本为「仅观测·挂窗待跑」
   临时基线,新基线本轮暂不锁定,建议以固定 brief+genre 口径正式重锚(_note 已记发现)。
2. trpg/puzzle 合成薄反例(黑箱点卡/合成薄壳)未入仓、本轮未跑。
3. 经营金标正例无仓内产物(golden-specs 只有 driver play-spec),待固化一款过门经营正例入仓。

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
 2026-07-03 15:25:43 -07:00
lili
d504e8f18a docs(boards): A波全落账——W-S1 修复三单✅+周边(genre路由/Service壳终结/预算两段式/gen.mjs守卫)、W-ARCH ②✅+判例库CI挂载✅(含ⓐⓑ扩容)、W-DSGN 文本批✅剩SVG、策划知识包待拍4件按创始人全权授权默认项拍定(ⓐ采纳落码/ⓑ冻结/ⓒ批入SoT/ⓔbacklog)、人办③改「复验在跑·绿即S0-S5·S6留创始人」
Some checks failed
contract-gates / contract-gates (push) Has been cancelled
Details
docs-gate / docs-gate (push) Has been cancelled
Details 
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
 2026-07-03 14:55:00 -07:00
lili
3495470d89 feat(quality): ⑨升通用底座第12条「核心操作非无脑」落 RICHNESS_CHECKLIST(L2 6→7,前11条一字不动,prompt 计数改动态 len 同源防漂)+ 测试分母随升;图说预算两段式「待实施/CLI仍hard」翻已落实态(cheap_budget 同源工厂+真M3实证) (质量SoT §4 规范一 12条口径,ba55508e 配套;金标复验=C波)
Some checks failed
contract-gates / contract-gates (push) Has been cancelled
Details
docs-gate / docs-gate (push) Has been cancelled
Details 
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
 2026-07-03 14:51:36 -07:00
lili
b78a8dfa4f merge: A1 核心修复批——单①驱动器约定收敛(C5接线:startRitual起局仪式+driver显式selectionBasis+sourceHash陈旧重生+金标held-constant不误杀)+单③反馈补厚(C6结构化单一来源:latch双支phaseNow+driverType+game-log摘要+CLI九门回喂对齐+tier2 sidecar富文本不变)+预算两段式(软停¥10/¥50只许收尾·on_acting拦生成面+硬地板×1.5 fail-closed,CLI/Service同源,W-ARCH②)+genre最小路由(五类词表·特异性降序)+Service壳合成终结事件+压缩崩fail-open补丁(SummarySchema全required根因坐实)。T7复现:fx4c-3 9/9·fx4c-1起局成功8/9;T8真M3 smoke:C6回喂环真转·软停真触发·硬地板真fail-closed(¥3.0158)
Some checks failed
contract-gates / contract-gates (push) Has been cancelled
Details
docs-gate / docs-gate (push) Has been cancelled
Details 
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
 2026-07-03 14:43:31 -07:00
lili
714d0adebc fix(cheap): judge_cheap_verdict 全绿路 feedback 文案改准(「九门全绿,放行」取代误导的「未定位到未过门」;passed=True 时 feedback 本不回喂,仅日志/trace 用) (T7 fx4c-3 复跑发现) 
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
 2026-07-03 14:32:57 -07:00
lili
51144f1a3b feat(cheap): T5 Service 壳合成终结事件(run 崩快失败,封 600s 慢失败出口)+ 压缩崩根因坐实与 fail-open 补丁 
- 压缩崩根因(venv 2.0.2 代码级证据,4 次实证 'important_discoveries' required):reply 每轮推理前
  compress_context(_agent.py:609)→ generate_structured_output 用 SummarySchema json-schema(五字段全
  required+maxLength,_config.py:11-53)强制 M3 tool_choice 产摘要 → dict 路 jsonschema.validate
  (model/_base.py:566-567)对缺字段/超长抛 ValidationError → 非 retryable 直接 raise(_base.py:426-428)
  → _compress_context_impl 非 overflow 支 raise e from None(_agent.py:461)→ 穿透 reply;ChatService.run
  吞异常不发终结事件(app/_service/_chat.py:166-176)→ driver 600s 慢失败
- ctx_compress_patch:fail-open 包 Agent.compress_context(压缩=旁路优化件,失败只跳过不杀主链;连续 2 次
  断路防每轮白烧压缩计费调用;钉 2.0.2 不改 venv 本体,纪律同 m3_stream_patch);Service 与 CLI 两处 apply
- collector on_reply 增崩溃支:先 flush 收口采集,再 yield 合成 ReplyEndEvent(消费侧先 publish 到 bus →
  driver 立刻收到回合终结按盘面快速失败),原异常原样重抛不掩盖;正常路零行为变化
- 测试 9 用例:成功透传/失败吞掉/2 次断路/成功清零/按实例隔离/apply 幂等;崩溃合成终结+异常上抛+崩溃路
  flush/正常路不追加/合成兜底不遮原异常

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
 2026-07-03 14:21:34 -07:00
lili
74c3ef51f8 docs(skills): puzzle skill §9 起局命中区教训条目对齐 v1.6.1 新约定(onEnter 纯几何建立、render 只画;保留原教训与实测证据)
Some checks failed
contract-gates / contract-gates (push) Has been cancelled
Details
docs-gate / docs-gate (push) Has been cancelled
Details 
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
 2026-07-03 14:20:45 -07:00
lili
d15119f3ad merge: A2 模板与提示词侧——6款模板起局按钮解耦render(纯几何函数+onEnter确定性赋值,no-render探针6/6进play+HEAD反事实6/6卡menu坐实)+ cheap-system v1.6.1 两条硬约束(targets键名固定/命中矩形onEnter建立)+cheap_roles逐字节parity+registry bump + gen.mjs入口守卫 + littlejs skill起局指导(卡menu案4修复面) 
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
 2026-07-03 14:20:08 -07:00
lili
d2b916f112 feat(cheap): T4 生产 create 路 genre 最小路由(brief→五类确定性词表·先验序=特异性降序·无命中回退通用) (M3 cutover 放行检查单增补) 
- cheap_genre_route:纯词表路由(零 LLM/零 IO/确定性),规则表数据化单源(_GENRE_RULES 顺序即先验序:
  非遗>TRPG>经营>解谜>剧情,特异词前泛词后);映射对齐 GENRE_BY_TEMPLATE 五类模板
- cheap_service_driver:取代硬编码 scaffold_template=None——命中选 per-genre 黄金骨架+genre 透传丰富度评分
  (生产 create 路与 S5 bake_off 复验同轨);无命中 (None,None) 走通用 _template(路由只加不减)
- tests/test_genre_route:五类命中/无命中回退/多类共现先验序/GENRE_BY_TEMPLATE 一致性/确定性/脏输入 14 用例

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
 2026-07-03 14:17:54 -07:00
lili
34923812c3 fix(gen): 模板起局鲁棒性 menuStartBtn 解耦 render + targets 键/起局约定进 prompt v1.6.1 + gen.mjs 入口守卫 (W-S1 修复单②) 
T1 六款模板(_template + feiyi/puzzle/shop/story/trpg)起局/重开按钮命中矩形:
从 render 副作用回写(menuStartBtn = renderMenu(...).startBtn)改为抽纯几何
menuStartRect/overRestartRect + 在场景进入点 onEnter 确定性赋值、render 只画;
治并发 rAF 饥饿下 render 未跑过=命中矩形恒 null=卡 menu(归因案4)。
no-render 探针坐实:修复前六款均卡 menu、修复后均按坐标起局进 play。

T2 cheap-system.md 升 v1.6.1,红线段增两条硬约束:①点 target 玩法 state 键名
固定为 targets(不得改成 items 等,C5 驱动器族选择依赖)②起局命中矩形须在状态
进入点建立、不依赖渲染帧;各模板 _forensicsView 加同义契约注释。cheap_roles.py
_SYSTEM_PROMPT 与 md 正文逐字节 parity 同步,registry.yaml 版本 bump 对齐。

T3 gen.mjs 顶层裸调 main() 加入口守卫(import.meta.url === pathToFileURL(argv[1])),
被 import 时不再触发真生成(footgun,曾 ~116s)、直接执行行为不变。

T4 puzzle-game-design.md §9 死链经核实在本 commit 不存在(唯一 play-spec 引用已
正确指向存在的 puzzle.play-spec.json),未伪造改动;littlejs-game-dev.md 补一条
起局 onEnter 指导闭合 doc-code。

验证:test_roles 12/12 + check_registry 17 对齐 + md↔_SYSTEM_PROMPT parity True
+ 6 模板 esbuild build 全绿 + amodel-gen tests 15+11 + 模板 node --test 除
_template-shop/headless-boot(既存红,与基线同)外全绿。

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
 2026-07-03 14:17:53 -07:00
lili
9e1e943503 feat(budget): T3 预算两段式统一(软停线 ¥10/¥50 只许收尾+RMB 硬地板 ×1.5 fail-closed 封上界+CLI/Service 同源工厂) (W-ARCH②,创始人 2026-07-03 裁决) 
- middleware:soft 档升两段式——越软停线置软停+on_acting 拦新增生成面工具(write_file/write_source/scaffold_init,
  收尾类 finish/check/build/跑门/只读恒放行);越硬地板(软停线×rmb_hard_floor_ratio=1.5)fail-closed 抛熔断,
  spent>floor 后 est≥0 使下次调用必拦 → 整局 ¥ ≤ floor+单笔,上界数学可证明;hard 档单段语义逐字不变
- genconfig+generation.yaml:rmb_hard_limit 3.0(spike 占位)→50.0(富档软停线裁决值)、新增 rmb_hard_floor_ratio=1.5
  / cheap_rmb_hard_limit=10.0 登内置默认,带裁决出处中文注释
- cheap_budget.build_cheap_breaker:便宜档 breaker 唯一装配点——CLI(cheap_studio,原 hard 单段 ¥10 改两段)与
  生产 Service(cheap_service_app,原 soft 无地板补地板)同源读 genconfig,档位参数不可被 overrides 改
- 测试:两段式 11 用例(软停区间放行/生成工具拒·收尾放行/硬地板到线即停/spent>floor 次调必拦/工厂同源 env 同变);
  test_budget_soft_stop 旧「soft 无界」口径按裁决更新为两段式并补越地板用例;既有 hard 档测试全量不动全绿

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
 2026-07-03 14:15:59 -07:00
lili
10de6aab8e merge: E14 图说 doc-sync 八条回同(配置反转/收敛环conditional/便宜档Python化+12插件/RepairMiddleware+预算两段式/checkpoint降准/DockerWorkspace句删/质量canonical互引/trace README口径)+ Δ6 接缝频率序 + Δ9① A8-A13 注记 + C5/C6 入契约表 + studio.py 头注释纠正(W-DSGN 文本批,SVG 留尾波)
Some checks failed
contract-gates / contract-gates (push) Has been cancelled
Details
docs-gate / docs-gate (push) Has been cancelled
Details 
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
 2026-07-03 14:12:44 -07:00
lili
6ef0b5d75e docs(gen-engine): 运行时图说 doc-sync 回写实现对账 12 项(W-DSGN) 
以 2026-07-03《生成引擎实现与图说一致性对账》§5 八条 + 重推演 Δ6/Δ9① +
预算两段式裁决为据,把掉队约两周的运行时图说追回现实,写前逐条重核工作树代码。

图说本体(主对象):
- 配置口径反转:ADR-4/A13/§5.2/§5.8/对标表/五列/§二补④ 从「Langfuse 式热取 + GitOps
  四闸」改为「yudao 配置中心版本化(MySQL 版本行)⊕ Nacos 下发 ⊕ AgentScope per-POST
  热重载」,阶段〇+一① 已代码落地;Langfuse 式退历史备选、原表保留作决策留痕。
- §4.2 收敛环:前瞻「go(留观微调)」改为真跑 conditional 结论,补 agent 层 win-balance
  瓶颈(约 1-2/5 thrash)、网关无强档记 conditional no-go、F-1 反馈补厚已落。
- §4.1 便宜档:补运行时载体已 Python 化(cheap-worker 双入口 CLI+Service、复用 tier2
  worker 包;Node 拆两半:gen.mjs 退对照、tools.mjs/serve-and-play.sh/play.cdp.cjs
  九门执行层 shell-out 仍活);「十一插件」核正为十二(8 基元+4 编排)。
- §5.3 续修:回同续修原语迁入 on_reasoning 的 RepairMiddleware(单 POST 内 finish 点
  续修=生产主路,外层 resume 为本地 runner 并存 fallback);预算两段式(软停线 ¥10/¥50
  + 硬地板 ×1.5=¥15/¥75、与轮数/墙钟任一先到即停),§5.2/§5.7 同步。
- §5.1 checkpoint 降准(每外层 resume 落本地 JSON;Redis durable 已编码未真跑);服务壳
  已编码真跑未验证、阶段一 Agent Team 已提前落地默认开启。
- §5.4 删 DockerWorkspace 前置阻断句(06-28 已裁 in-process);A2.5 表行改现·部分;
  补引质量 canonical(D11 权重管辖已移交、四层塔与三层校验共用 L1-L3 但异轴防混)。
- §一 Δ6：第一原则改述为「接缝按变更频率设」(模型月换>品类周扩>引擎季增>框架可能永不换;
  协议脊柱与 judge 独立两资产保留;框架适配验证明确为远期非投入项)。
- §三 Δ9①：A8–A13 加「单实现期不冻、第二消费者出现才抬升」注记(不删行);C5/C6 判分
  闭环两契约入表(PlaySpec 考卷/VerdictFeedback 判卷反馈,已立 schema+校验器、生产接线
  在途,指 contracts/play-loop/)。
- A11(工单第 10 条)复核:图说 §六已是完整 A11 章节，无需重复补——系对账/工单误判。

配套一处:contracts/trace/README.md 廉价线口径 SAA→cheap-worker(便宜档已 Python 化复用
tier2 TraceAdapter/schema,saa-trace-event schema 保留作 SAA 远期轨立位)。
代码触点一处:tier2 agent_loop/studio.py 头注释纠正（「不再外层 repair」与同文件外层
resume 相左）。
收尾:两份配置设计档 sot-impact「收口时回写」翻「已回写」；plan① §11 TODO ⑤⑥ 勾状态。

docs-gate 七检全绿；studio.py 语法校验通过。

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
 2026-07-03 14:10:56 -07:00
lili
93cdffa33e feat(cheap): T2 续修反馈补厚·C6 判卷契约接线(latch 双支 phaseNow+driverType+game-log 摘要+CLI 九门回喂对齐+tier2 结构化 sidecar) (W-S1 修复三单·单③) 
- gate_judge:反馈先构造 C6 结构化对象(build_cheap_feedback_obj)再渲染中文(单一来源,拼串丢字段结构上不可能);
  latch advisory 支(phaseNow)与驻留支(after)都读;每条逐门带 driverType(staged play-spec 取数);
  game-log.json 告警/报错摘要进反馈(截断保护);C6 对象落 evidence/verdict-feedback.json 证据留痕
- cheap_studio:CLI 收敛判据升「九门绿」——finish 后 _closeout_gates 跑门,未绿且轮数/预算有余带 C6 反馈
  resume 续修(与生产 RepairMiddleware finish 点拦截行为同型);软停后只收尾不回喂;收口复用最后一轮门结果;
  对照路 run_gates=False 零改动
- tier2 对齐(有界零行为变更):verdict_feedback 富文本逐字不动,新增 verdict_feedback_obj+write_feedback_sidecar
  结构化 sidecar 落 evidence/(judge_tier2_verdict 未过时 best-effort)
- Service _cheap_check / driver _build_summary 传 game_id+staged_dir 解锁完整反馈口径
- tests/test_feedback_c6.py:phaseNow 两支/负样本被 validate.py 拦/driverType/game-log 截断/sidecar/富文本不变 14 用例

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
 2026-07-03 14:08:44 -07:00