Commit Graph

9237 Commits

Author SHA1 Message Date
jxxghp
5e50caabcb chore(architecture): sync ruff low watermark 2026-09-11 11:16:01 +08:00
jxxghp
92d809f351 refactor(api): isolate module endpoints 2026-09-11 11:08:03 +08:00
jxxghp
895fda2a01 chore(architecture): refresh module dependency baseline 2026-09-11 10:49:08 +08:00
jxxghp
ad73a14502 test(api): refresh generated contract baselines 2026-09-11 10:39:28 +08:00
jxxghp
ae1187bd3f feat(module): add dynamic module switches and catalogs 2026-09-11 10:31:53 +08:00
jxxghp
0e26fe0dc9 eval: align native collaboration discovery 2026-09-11 10:23:26 +08:00
jxxghp
cca0dd4a6c eval: add Codex OAuth parity comparison 2026-09-11 09:50:56 +08:00
jxxghp
556543023f eval: switch live benchmark to native Gemini 3 2026-09-11 08:41:10 +08:00
jxxghp
27cb5d94ba docs(agent): record skill loading and input feedback evidence 2026-09-11 08:08:30 +08:00
jxxghp
0f8aa4dfa9 fix(agent): return operation input contracts on errors 2026-09-11 08:07:54 +08:00
jxxghp
13b4fe1128 merge: sync remote skill loading optimization 2026-09-11 07:54:02 +08:00
jxxghp
3a22cb0118 Revert "fix(agent): surface common API field contracts"
This reverts commit bdeef730b9.
2026-09-11 07:54:02 +08:00
jxxghp
f5bc72ff44 fix(agent): load split skill docs through read_skill 2026-09-11 07:50:18 +08:00
jxxghp
bdeef730b9 fix(agent): surface common API field contracts 2026-09-11 07:50:13 +08:00
jxxghp
c5a5eb0da0 feat(evaluation): capture tool catalog implementation evidence 2026-09-11 07:47:19 +08:00
jxxghp
b4cf40d46a Merge remote-tracking branch 'origin/v3' into codex/agent-evaluation-20260909
# Conflicts:
#	skills/moviepilot-api/api/download.md
2026-09-11 07:32:54 +08:00
jxxghp
e288222bff docs(agent): record latest Gemini evaluation evidence 2026-09-11 07:30:42 +08:00
jxxghp
313ce80f58 test(agent): update skill operation count 2026-09-11 07:29:58 +08:00
jxxghp
a731cc0ce6 docs(agent): record Gemini evaluation and harness boundaries 2026-09-11 07:22:37 +08:00
jxxghp
c5d5698e04 feat(agent): complete moviepilot api contracts 2026-09-11 07:21:05 +08:00
jxxghp
8c0746172a docs(agent): make download contract self contained 2026-09-11 07:13:46 +08:00
jxxghp
7195a29c33 fix(evaluation): scope skill file reads to temp agent 2026-09-11 07:08:48 +08:00
jxxghp
a2f40ceecc refactor(agent): expose split skill files in evaluation 2026-09-11 07:03:12 +08:00
jxxghp
1eda214dfc docs(agent): clarify common API contracts 2026-09-11 06:53:44 +08:00
jxxghp
74db54f067 refactor(agent): keep one general subagent profile 2026-09-11 06:48:00 +08:00
jxxghp
dbd0d68f8f docs(refactor): move design documents 2026-09-11 06:29:12 +08:00
jxxghp
7ab8cddccb feat(evaluation): support chat completions providers 2026-09-11 06:26:26 +08:00
jxxghp
1a1aa59838 docs(architecture): move refactor documents 2026-09-11 06:20:59 +08:00
jxxghp
b1639b25b6 docs(agent): record general subagent rollout 2026-09-11 06:10:53 +08:00
jxxghp
00150bb87f refactor(agent): default delegation to general subagent 2026-09-11 06:09:04 +08:00
jxxghp
fb6859cc11 docs(agent): record latest provider evaluation round 2026-09-11 05:56:21 +08:00
jxxghp
72aabe12f9 refactor(agent): keep WebAgent stream within complexity budget 2026-09-11 05:52:09 +08:00
jxxghp
b55a02519b chore(architecture): record agent module dependency edges 2026-09-11 05:43:02 +08:00
jxxghp
34ec415d90 docs(agent): record final provider evaluation evidence 2026-09-11 05:24:10 +08:00
jxxghp
abd3cea92b feat(agent): queue steering input during active runs 2026-09-11 05:17:37 +08:00
jxxghp
66d288ffc2 feat(agent): isolate terminal task ownership 2026-09-11 03:12:16 +08:00
jxxghp
03ec668fe3 test(runtime): align strict event mocks 2026-09-11 00:20:57 +08:00
jxxghp
d32d17568a fix(runtime): apply configuration changes without restart 2026-09-11 00:12:05 +08:00
jxxghp
ad2da6eb8b test(agent): cover split api skill support files 2026-09-10 23:09:02 +08:00
jxxghp
d096a1380d docs(agent): split moviepilot api skill contracts 2026-09-10 23:00:16 +08:00
jxxghp
f383a8a5a3 feat(agent): 接入媒体自动分类策略 API 2026-09-10 22:36:50 +08:00
jxxghp
8c3a394518 docs: 同步优化清单中全量 mypy 历史债务指标 2026-09-10 20:48:36 +08:00
jxxghp
f14601eb9a chore(ratchet): 同步 add_download 降低的 mypy 类型错误基线 2026-09-10 20:41:46 +08:00
jxxghp
171cc9b973 fix(schemas): 规范 import 格式并简化 MetaInfo 属性定义 2026-09-10 20:33:47 +08:00
jxxghp
e857e2cf04 fix(schemas): 保持内部导入为私有,避免 MetaBase.to_dict 导出无 setter 属性 2026-09-10 20:23:31 +08:00
jxxghp
60a95090b6 fix(workflow): 工作流添加下载支持 MetaInfo.season_list (#6637) 2026-09-10 20:13:04 +08:00
jxxghp
0192669077 Merge pull request #6639 from thsrite/fix/mediavault-review-findings 2026-09-10 19:43:25 +08:00
thsrite
2b4ba9c927 fix(mediaserver): 翻页按总数收敛,剧集定位命中即返回
按代码审查意见再修两处,都与上一轮引入的关键字翻页有关:

- 末页恰好装满时仍会多请求一页。响应里已带可靠总数,改为累计条数达到总数即
  结束,只有没有总数时才按页是否满额判断;那次多余请求一旦失败,会把完整结果
  误报成服务不可达
- 按标题定位剧集先把所有页读完再检查匹配,第一页已命中的目标会因后续页请求
  失败而返回不可达,比改动前更差;改为逐行检查、命中即返回,只有在命中前中断
  才报告不可达

补 2 条回归,均已确认在修正前失败。
2026-09-10 17:59:08 +08:00
thsrite
963a80788e fix(mediaserver): 翻页中途失败不再被当成查询结束
新增的关键字翻页逻辑只在第一页失败时返回 None,后续页面请求失败会让生成器
正常结束:get_movies 返回残缺列表、按标题定位剧集返回空串,等于把「连不上」
误报成「确定不在库中」,可能触发重复下载。

改为中途失败抛出私有异常并由调用方转成 None,恢复「连接失败返回 None、
未找到返回空结果」的区分;补 2 条回归,均已确认在修正前失败。
2026-09-10 17:35:27 +08:00
thsrite
429b0020be fix(mediaserver): 修正 MediaVault 图片凭据、查询截断、剧集类型与刷新短路
代码审查在 v2 移植版上发现四处缺陷,同样存在于本分支已合入的模块:

- 图片地址把管理面板的长效 API Key 拼进查询参数。该地址会随接口响应交给浏览器
  直接加载,等于把管理凭据下发到客户端,可被重放到 MediaVault 的全部管理接口。
  MediaVault 的媒体库图片为免鉴权资源,去掉该参数即可
- get_movies 与按标题定位剧集只读关键字查询的第一页,命中数超过单页上限时排在
  后面的目标会被误判为未入库;改为翻页直到取完
- 最新入库把 Series 条目标成电影,只有 Episode 被识别为电视剧;改为两者都映射为
  电视剧,「季:集 - 分集名」副标题仍只给分集
- refresh_library_by_items 把生成器交给 all(),首个媒体库排队失败就会短路,同批
  命中的其余媒体库收不到扫描任务;改为先全部发出请求再汇总

补 6 条回归测试;同时收紧测试替身:假 Api 改为复用真实的 image_url 实现,避免
自行拼装掩盖「图片地址不得携带凭据」这一约束。
2026-09-10 17:28:58 +08:00