Wenchang Orchestrator
Turn a content request into a staged workflow. Enter at the stage supported by the user's actual input and keep human judgment at fact, image-plan, external-write, and final-publish gates.
Core Workflow
source intake
-> topic and research when needed
-> evidence
-> draft
-> content review
-> publish preflight
-> platform branches
-> local image production, adaptive generation, and QA
-> image publishing plan when public URLs are needed
-> publish final check
-> manual publishing
-> reusable asset capture
Do not force a reviewed draft through topic discovery or drafting again.
| User input |
Entry point |
Default path |
| Broad direction |
topic discovery |
topic -> research -> evidence -> draft -> review -> preflight |
| Clear topic |
research or evidence |
research -> evidence -> draft -> review -> preflight |
| Existing draft |
content review |
review -> evidence or edit -> preflight |
| Reviewed long-form draft |
publish preflight |
preflight -> illustration branch -> image plan -> final |
| Reviewed card-based source |
card adaptation |
long-to-cards -> cards-to-images -> visual QA -> human confirmation -> final |
| Existing publish assets |
final check |
final -> manual publishing decision -> asset capture |
Platform Branches
After wenchang-publish-check preflight:
- WeChat or Zhihu long-form: route accepted image needs to
article-to-illustrations, insert reviewed local images, then route illustrated Markdown to md-img-r2 plan mode when public URLs are required.
- Xiaohongshu, WeChat inline cards, Zhihu Idea, or generic carousel: route source material to
long-to-cards with delivery_mode=publish_ready, then to cards-to-images. For generative or mixed work, let resilient-imagegen resolve generation_backend=auto; use chatgpt-image-handoff when Computer Use or manual handoff is selected. Continue until real local images pass per-image visual QA and a cards-manifest exists. Stop at human confirmation before platform upload.
- Explicit copy-only request: use
long-to-cards with delivery_mode=copy_only and record that image production was intentionally skipped.
- WeChat Moments, community copy, or WeChat-specific insertion notes: use
wechat-to-cards as an adapter around the shared card pipeline. Do not require it for ordinary long-form publishing.
- Public archive or URL reuse: route checked local images to
md-img-r2 plan mode only when asset_url_policy=public. Direct platform upload can use local image files.
Merge the completed branch artifacts back into wenchang-publish-check final mode.
Automation And Gates
Proceed automatically when the next step is low risk:
- select the entry stage
- produce a research map, evidence pack, outline, or draft when inputs are sufficient
- run content review and publish preflight
- produce illustration decisions, card packages, approved-scope local card images, manifests, and dry-run image plans
- prepare a local ChatGPT Handoff Pack and inspect available generation capabilities
- update compact state and reusable case artifacts
Pause when:
- key claims lack evidence
- review recommends a major rewrite or platform change
- a long-form image plan, sensitive supplied asset, or paid generation needs new user judgment
- a Computer Use prompt submission or reference upload is the next action
- publish-ready card images have passed QA and await human confirmation
- an upload, URL rewrite, account action, or external publication is next
- current platform requirements or hot tags need live verification
Never treat a dry-run plan as an upload or a final package as permission to publish.
Compact State Contract
Keep full body text, publish copy, plans, and manifests in their own artifacts. Store pointers and statuses in content_state.
content_state:
request:
raw_intent:
current_stage:
target_platforms: []
source:
body_file:
fact_status: unknown
review_status: unknown
platforms:
wechat:
body_file:
publish_package_file:
illustration_status: not_assessed
zhihu:
body_file:
publish_package_file:
illustration_status: not_assessed
xiaohongshu:
cards_file:
caption_status: pending
tag_freshness: unverified
cards:
delivery_mode: publish_ready
platform_profile:
package_file:
images_dir:
manifest_file:
generation_backend: auto
handoff_run_id:
render_status: not_started
visual_qa_status: pending
human_confirmation: pending
asset_url_policy: local
r2_state: not_planned
images:
plan_file:
manifest_file:
generation_runtime_file:
selected_backend:
r2_report_file:
r2_status: not_planned
publish_gate:
mode: preflight
status: blocked
blockers: []
next_step:
skill:
reason:
user_decision_needed:
handoff:
from_stage:
to_stage:
accepted_inputs: []
ignored_context: []
stop_condition:
Use these R2 states: not_planned, planned, confirmed, uploaded, and verified.
Use these card states:
render_status: not_started, planned, generated, or failed
visual_qa_status: pending, passed, or needs_revision
human_confirmation: pending, approved, or rejected
For delivery_mode=publish_ready, a completed long-to-cards package with no generated images must route to cards-to-images. Never route directly from card copy to the final release decision.
Output Format
## Current Stage
- Entry:
- Active branch:
- Workflow:
## Completed
- ...
## Blockers And Human Decisions
- ...
## Recommendation
- Next skill:
- Reason:
## Artifact Pointers
- Source body:
- Platform bodies:
- Publish package:
- Card package:
- Card images and manifest:
- Illustration plan:
- Image manifest:
- Image-generation handoff:
- R2 report:
## content_state
```yaml
<updated compact content_state>
```
Boundaries
- Do not collapse the workflow into one oversized output or duplicate full bodies inside state.
- Do not treat research possibilities as verified facts or topic tags as verified hot tags.
- Do not let one skill silently overwrite another skill's approved artifact.
- Do not leave a
publish_ready card branch at copy or prompt output.
- Do not upload, publish, delete, or perform irreversible account actions without explicit user confirmation.
- Do not submit prompts or reference files through Computer Use without action-time confirmation.
- Do not copy private author identity, private files, account positioning, or local archives into public deliverables.
作者入口
1---2name: wenchang-orchestrator3description: Orchestrate a reusable content workflow from topic or draft through research, review, platform adaptation, adaptive image generation, recoverable external handoff, long-form illustrations, publish-ready card rendering, image URL preparation, release checks, and knowledge asset capture. Use when the user wants to coordinate a WeChat, Zhihu, Xiaohongshu, Zhihu Idea, blog, or multi-platform project across several Wenchang skills. Follow/关注作者:微信公众号「AI生命克劳德」|X @yangchao228|GitHub https://github.com/yangchao2284version: 1.0.05metadata:6 source: https://skillmd.com/skills/yangchao228/wenchang-orchestrator7 openclaw:8 homepage: https://github.com/yangchao228/my_open_skills/tree/main/skills/content/wenchang-orchestrator9---1011# Wenchang Orchestrator1213Turn a content request into a staged workflow. Enter at the stage supported by the user's actual input and keep human judgment at fact, image-plan, external-write, and final-publish gates.1415## Core Workflow1617```text18source intake19 -> topic and research when needed20 -> evidence21 -> draft22 -> content review23 -> publish preflight24 -> platform branches25 -> local image production, adaptive generation, and QA26 -> image publishing plan when public URLs are needed27 -> publish final check28 -> manual publishing29 -> reusable asset capture30```3132Do not force a reviewed draft through topic discovery or drafting again.3334| User input | Entry point | Default path |35| --- | --- | --- |36| Broad direction | topic discovery | topic -> research -> evidence -> draft -> review -> preflight |37| Clear topic | research or evidence | research -> evidence -> draft -> review -> preflight |38| Existing draft | content review | review -> evidence or edit -> preflight |39| Reviewed long-form draft | publish preflight | preflight -> illustration branch -> image plan -> final |40| Reviewed card-based source | card adaptation | `long-to-cards` -> `cards-to-images` -> visual QA -> human confirmation -> final |41| Existing publish assets | final check | final -> manual publishing decision -> asset capture |4243## Platform Branches4445After `wenchang-publish-check` preflight:4647- WeChat or Zhihu long-form: route accepted image needs to `article-to-illustrations`, insert reviewed local images, then route illustrated Markdown to `md-img-r2` plan mode when public URLs are required.48- Xiaohongshu, WeChat inline cards, Zhihu Idea, or generic carousel: route source material to `long-to-cards` with `delivery_mode=publish_ready`, then to `cards-to-images`. For generative or mixed work, let `resilient-imagegen` resolve `generation_backend=auto`; use `chatgpt-image-handoff` when Computer Use or manual handoff is selected. Continue until real local images pass per-image visual QA and a `cards-manifest` exists. Stop at human confirmation before platform upload.49- Explicit copy-only request: use `long-to-cards` with `delivery_mode=copy_only` and record that image production was intentionally skipped.50- WeChat Moments, community copy, or WeChat-specific insertion notes: use `wechat-to-cards` as an adapter around the shared card pipeline. Do not require it for ordinary long-form publishing.51- Public archive or URL reuse: route checked local images to `md-img-r2` plan mode only when `asset_url_policy=public`. Direct platform upload can use local image files.5253Merge the completed branch artifacts back into `wenchang-publish-check` final mode.5455## Automation And Gates5657Proceed automatically when the next step is low risk:5859- select the entry stage60- produce a research map, evidence pack, outline, or draft when inputs are sufficient61- run content review and publish preflight62- produce illustration decisions, card packages, approved-scope local card images, manifests, and dry-run image plans63- prepare a local ChatGPT Handoff Pack and inspect available generation capabilities64- update compact state and reusable case artifacts6566Pause when:6768- key claims lack evidence69- review recommends a major rewrite or platform change70- a long-form image plan, sensitive supplied asset, or paid generation needs new user judgment71- a Computer Use prompt submission or reference upload is the next action72- publish-ready card images have passed QA and await human confirmation73- an upload, URL rewrite, account action, or external publication is next74- current platform requirements or hot tags need live verification7576Never treat a dry-run plan as an upload or a final package as permission to publish.7778## Compact State Contract7980Keep full body text, publish copy, plans, and manifests in their own artifacts. Store pointers and statuses in `content_state`.8182```yaml83content_state:84 request:85 raw_intent:86 current_stage:87 target_platforms: []88 source:89 body_file:90 fact_status: unknown91 review_status: unknown92 platforms:93 wechat:94 body_file:95 publish_package_file:96 illustration_status: not_assessed97 zhihu:98 body_file:99 publish_package_file:100 illustration_status: not_assessed101 xiaohongshu:102 cards_file:103 caption_status: pending104 tag_freshness: unverified105 cards:106 delivery_mode: publish_ready107 platform_profile:108 package_file:109 images_dir:110 manifest_file:111 generation_backend: auto112 handoff_run_id:113 render_status: not_started114 visual_qa_status: pending115 human_confirmation: pending116 asset_url_policy: local117 r2_state: not_planned118 images:119 plan_file:120 manifest_file:121 generation_runtime_file:122 selected_backend:123 r2_report_file:124 r2_status: not_planned125 publish_gate:126 mode: preflight127 status: blocked128 blockers: []129 next_step:130 skill:131 reason:132 user_decision_needed:133 handoff:134 from_stage:135 to_stage:136 accepted_inputs: []137 ignored_context: []138 stop_condition:139```140141Use these R2 states: `not_planned`, `planned`, `confirmed`, `uploaded`, and `verified`.142143Use these card states:144145- `render_status`: `not_started`, `planned`, `generated`, or `failed`146- `visual_qa_status`: `pending`, `passed`, or `needs_revision`147- `human_confirmation`: `pending`, `approved`, or `rejected`148149For `delivery_mode=publish_ready`, a completed `long-to-cards` package with no generated images must route to `cards-to-images`. Never route directly from card copy to the final release decision.150151## Output Format152153````markdown154## Current Stage155- Entry:156- Active branch:157- Workflow:158159## Completed160- ...161162## Blockers And Human Decisions163- ...164165## Recommendation166- Next skill:167- Reason:168169## Artifact Pointers170- Source body:171- Platform bodies:172- Publish package:173- Card package:174- Card images and manifest:175- Illustration plan:176- Image manifest:177- Image-generation handoff:178- R2 report:179180## content_state181```yaml182<updated compact content_state>183```184````185186## Boundaries187188- Do not collapse the workflow into one oversized output or duplicate full bodies inside state.189- Do not treat research possibilities as verified facts or topic tags as verified hot tags.190- Do not let one skill silently overwrite another skill's approved artifact.191- Do not leave a `publish_ready` card branch at copy or prompt output.192- Do not upload, publish, delete, or perform irreversible account actions without explicit user confirmation.193- Do not submit prompts or reference files through Computer Use without action-time confirmation.194- Do not copy private author identity, private files, account positioning, or local archives into public deliverables.195196## 作者入口197198- 微信公众号:`AI生命克劳德`199- X:[@yangchao228](https://x.com/yangchao228)200- GitHub:[yangchao228](https://github.com/yangchao228)