loopilot-workbench
Skill1 starsChanged 27 days agoApache-2.0View on GitHub
Run and open the local LooPilot operations workbench in Codex Browser. It indexes real report manifests and SQLite state, exposes safe local media, stores auditable human content reviews and publish-readiness facts, and lets Codex materialize approved reviews for workflow resume without requiring Feishu.
In this repository10 other files it gives its agents.
- content-reviewerSkill
- creative-runnerSkill
- douyin-music-librarySkill
- ecommerce-operationsSkill
- loopilot-deployerSkill
- loopilot-doctorSkill
- reusable-material-librarySkill
- video-content-understandingSkill
- video-editor-runnerSkill
- video-script-generatorSkill
The file
We split this skill into the rules it gives an agent
4 rules. About 0 are common, 2 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 8
Use this skill to start or reuse the local LooPilot workbench and show the current foreground business task in the current Codex chat's …
- Very specific: line 77
Every review is append-only.
- Distinctive: line 79
Run the real local HTTP bridge smoke without external side effects:
- Distinctive: line 105
Do not stop the service merely because the current Codex task finished.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
LooPilot Workbench
Use this skill to start or reuse the local LooPilot workbench and show the current foreground business task in the current Codex chat's right panel. Follow 共同任务约定; opening the workbench itself does not create a business task.
默认入口:新版深色工作台
所有正式前台 LooPilot 技能任务,包括商品、创作、投放、运营、查询和诊断,默认展示新版深色工作台。使用 manage_merchant.py,不以旧版 manage_workbench.py 的页面作为默认结果页。
先沿运行配置及商家 state_root/service.json 确认原服务和商品库;端口采用返回值,不硬编码或把旧版端口当新版。用实际目录执行:
python3 skills/loopilot-workbench/scripts/manage_merchant.py ensure-running \
--state-root /absolute/user/state/merchant-workbench \
--catalog-root /absolute/user/state/product-catalog \
--path /overview核对 GET /api/health 的 service=loopilot-merchant-workbench、state/catalog/code 根。服务未运行时按已有 service.json/配置补齐原启动参数再启动,不创建空商品库或丢弃素材/投放配置。仅有旧版程序时记录新版安装缺口,不启动旧版充当新版。
- 已有商家任务:将
--path替换为--task-id TASK --shop SHOP,由真实任务投影返回精确详情 URL,不猜任务身份。 - 无商家任务:新版
/overview,并在对话提供原结果文件;通用 bridge 的workbench_path是新版展示路径,result_file是结果文件。旧task_path=/runs/...只用于兼容报告,不能作为默认打开目标,也不能伪造成商品任务。新版总览不表示通用任务已进入四类商品任务列表。 - 主 Agent 用
open_in_codex在当前聊天右侧打开返回 URL,省略threadId;已有本聊天 IAB 工作台标签时复用并导航,旧版页也在同一标签切到新版。URL 已匹配时保持;系统 Chrome 页不等于侧栏,queued不等于打开成功,需回读实际 URL/页面。 - 开始与完成均沿此入口;子阶段只更新原记录。后台采集不打开/切换页面。展示暂不可用时保存准确 URL 和原因,继续可执行部分,不改业务状态或自动回退旧版。
商家独立工作台
商家商品库与四类任务(视频生成、标题优化、图片优化、投放优化)读取 商家工作台入口;应用、每日采集、七天结论及历史补录读取 任务跟踪操作参考。独立服务使用 manage_merchant.py,返回实际详情 URL;旧版达人报告入口仅在明确请求时使用。页面与对话共用同一任务记录,网页保存不启动生成或线上应用。
任务统一显示待开始、执行中、待处理、已完成、已发布,废弃单列筛选。已完成代表产出完成,人工审核和发布授权沿对话;发布/应用成功由真实回执派生,具体查询、返修和版本规则见任务跟踪。
新版商家投放任务复用 ad_optimization,以 options.ad_category=budget|material 区分预算与素材调优;预算任务作用于同店、同账户、同场景的一个计划,多商品只关联展示。素材任务关联原图片/视频创作任务,审核事实沿原 human_review_store,任务状态注记不能代替审核。视频方案的用户启动确认与成品审核分别展示和回读;页面不直接生成、上传或修改投放。执行说明见 千川自动调优。
旧版兼容能力(仅明确请求旧版时使用)
- Pages:
overview,products,creative,creative/<creative_id>,publishing,videos, andruns/<run_id>. - Navigation: the main menu stays in a sticky top header at every viewport width; detail pages return to their parent business page, with run parents resolved from
run.module. - Runtime: Python standard library plus native HTML/CSS/JavaScript.
- Service: loopback-only, on-demand, persistent across Codex tasks, and explicitly stoppable.
- Data: production mode discovers validated
report-manifest-v1files only under configured run roots, with optional explicit recent-day legacy backfill. The SQLite index is disposable and rebuildable; run artifacts remain the truth source. - Accounts: list pages aggregate all accounts by default, expose a per-page account filter, display the human-readable account name, and retain
account_keyplusaccount_idfor stable isolation. - Creative media: the list and detail pages expose only task-declared media through opaque media IDs; video playback supports HTTP Range.
- Remix tasks: 混剪与普通创作共用
creative页面,以task_type=remix/creation标识和筛选;详情可查看方案 revisions、实际使用原素材和成片。素材入库进度不进入页面,用户通过 Codex 按导入任务 ID 查询。 - Formal batch download:
/api/creative-tasks/formal-batch-download只打包用户勾选且formal_publish_ready=true的 MP4,使用客户/商品/任务/变体/方案 revision 命名临时 ZIP;请求结束后清理临时文件,不代表发布授权。 - Video replication: detail pages read back the exact
replication_modeandgeneration_strategy; strong-fidelity jobs display “逐镜强保真” and “先锁画面,再统一配音” without enabling generation or publish actions. - Local write boundary: publish readiness plus append-only
human-reviews/events.jsonland its atomiccurrent.jsonprojection understate_root. An approved workflow review may be materialized into that workflow'sreviews/current.jsonandmanual_creative_review_snapshot.jsonfor Codex/heartbeat readback. - Safety: reviews and publish readiness do not request or execute generation, editing, publishing, cart changes, ad spend, configuration writes, or Feishu writes.
原创任务
原创任务复用 /creative 及原详情页,task_type=creation / creation_mode=product_original,任务类型显示“原创”。同页读取完整方案、商品事实、主档进度/额度、实际商品/主体/场景素材和成片;未知费用显示未知,不以剩余额度代替已用费用。无参考视频不显示“参考缺失”。不新增入口或网页生成/重试/改方案接口,原创详情不提供发布准备操作。
原生 report producer 可从任务文件零 Provider 重建投影;业务完成以当前主档 final result 与实际文件 SHA 为准,不以 Provider 成功或任意 MP4 存在代替。缓存非权威,媒体越界/串账号不开放。
显式查看旧版报告
以下命令仅在用户明确要求旧版达人/报告页面或开发演示时使用,不作为正式任务完成后的默认展示入口:
python3 skills/loopilot-workbench/scripts/manage_workbench.py ensure-running \
--run-root /absolute/user-owned/loopilot/runs \
--path '/runs/RUN?module=MODULE&account_key=ACCOUNT'沿原报告完整 module/account_key/run_id 身份查询,不能仅用 run 名称。旧服务的 /overview、/runs/... 与新版不是同一页面,不能只替换端口。旧服务与数据继续保留,但新版展示失败不自动回退到旧版。
演示数据仅允许显式开发 --demo-mode,不能替代生产;--backfill-days N 只用于用户明确要求的只读历史补扫。
Human review and Codex bridge
Reviewable manifest assets use stages product_image, model_image, video_prompt, raw_video, or final_video, with an immutable asset SHA256, payload SHA256, object revision, account and workflow ID. The API derives those fields from the manifest; the browser may submit only decision, comment, reviewer and matching account.
Every review is append-only. Asset, prompt payload, account, object or revision changes make the old decision stale. approved may be materialized for the control plane, but the workbench never calls the next execution skill itself. rejected, changes_requested, stale, missing or cross-account decisions remain blocked.
Run the real local HTTP bridge smoke without external side effects:
python3 skills/loopilot-workbench/scripts/smoke_review_bridge.py \
--output-root /absolute/user-owned/loopilot/state/workbench-smokeExecution skills now write lifecycle manifests natively. Verify the complete creative/review/edit/local-publish projection without network or external writes:
python3 skills/loopilot-workbench/scripts/smoke_specialist_reporting.py \
--output-root /absolute/user-owned/loopilot/state/specialist-report-smokeProduction product and creative views merge summary.product, summary.creative, and summary.publishing by exact account and product identity. A ready or published fact may not be downgraded by an older draft.
See references/report-contract.md for manifest, media and review contracts.
旧版服务生命周期
python3 skills/loopilot-workbench/scripts/manage_workbench.py status
python3 skills/loopilot-workbench/scripts/manage_workbench.py restart --path /overview
python3 skills/loopilot-workbench/scripts/manage_workbench.py stopDo not stop the service merely because the current Codex task finished. Heartbeat and background tasks may refresh data later, but must not automatically expand the Browser sidebar.
Truth-source boundary
Local business run artifacts remain the intended truth source. Feishu integration status and outbox counts are display-only metadata and never overwrite local business facts.
Changing the selected publish account, work title, video task, or video content invalidates the previous ready state. Publishing and published tasks are immutable in the workbench; the publisher must still perform its own fresh readback, explicit confirmation, platform action, and result verification.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

