Connect account positioning, content goals, evidence, platform-specific expression, visual assets, publishing, and performance review into a resumable closed loop. Prioritize trustworthy, platform-native, actionable content rather than treating mass generation as a substitute for genuine judgment.

Determine the content root directory in this order:

Do not assume that a particular Vault, account, growth plan, or vendor tool exists. Save text deliverables in Markdown by default unless the project rules or the user explicitly require another format.

Load only the modules necessary for the outcome the user actually needs:

self-media-content-brief

self-media-platform-copywriting

self-media-content-strategy

self-media-trend-radar

self-media-short-video

self-media-video-publisher

self-media-content-analytics

self-media-content-delivery

self-media-wechat-publisher

Before using a sub-Skill, read its entire . If a module is not installed, run the minimum workflow for the corresponding stage of this Skill; do not invent capabilities that do not exist.

Requirement clarification → Awaiting direction confirmation → Research and evidence → Awaiting platform confirmation → Publishing preflight → In creation → Awaiting title confirmation → Asset generation → Quality review → Awaiting final-draft approval → Pending publication → Draft created or published → Pending performance review → Performance review completed

The following transitions must be completed by the user:

Final-draft approval does not imply publishing authorization.

Invoke . Consolidate the account positioning, target audience, content goals, core thesis, evidence, tone, depth, exclusions, and platform angles into a single confirmation card. If information is insufficient, ask no more than three high-impact questions and allow the user to accept reasonable defaults.

Invoke as needed. Prioritize the user's real experiences, backend data, original assets, official documentation, and first-party publications. Use competitors only to identify demand, structure, and opportunities for expression; do not copy their content.

The evidence package must contain at least:

Read platforms.md. Recommend one primary platform and no more than two expansion platforms. Explain each platform's role, deliverables, and time cost.

Read preflight-and-registry.md. Check only the capabilities required by the selected platforms: copy, images, raw video or screen recordings, sensitive digital-human assets and platform readiness, video compositing and subtitles, draft creation, and analytics. Inspect the animation production pipeline and music licensing only when the video production mode is animated-explainer production, and inspect video uploading only when the user wants the workflow to upload videos. Inspect sensitive digital-human assets and platform readiness, as well as video compositing and subtitles, only when the video production mode is digital-human production. For other content types and tasks involving a live presenter or asset analysis, mark the digital-human readiness and video compositing or subtitle checks as not applicable, and do not enter solely for that reason. When a capability is unavailable, provide a fallback such as a manual publishing package. Rights confirmation, manual uploading, third-party data processing, generation, compositing, and acceptance criteria for digital-human production are covered in digital-human-production.md.

Invoke . Design the title, opening, evidence order, structure, and call to action separately for each platform; do not mechanically shorten one version.

Internally generate 6 to 8 titles covering angles such as outcomes, benefits, identity, conflict, contrast, curiosity, and numbers. After filtering, show only the three strongest to the user and explain the angle that best fits each one.

Save the complete first draft before waiting for title confirmation. Do not generate a cover before the title is confirmed.

Generate assets only after the title is confirmed. First, recommend 2 to 3 styles from the visual style library () based on the content type and platform, explain the differences in one sentence each, and wait for the user to choose. If the account already has a stable style preference, use it by default and notify the user. If the module is not installed, confirm the art style, color palette, and composition directly with the user.

Record the selected style in the task card's visual anchor, then create a copy-to-visual mapping. For software or product content, use real interface assets when available; otherwise, use scenes or metaphors. Do not fabricate product UI.

Invoke for video. By default, when no source footage is available, deliver the script, storyboard, subtitles, cover, and publishing fields. If the user explicitly selects digital-human production, portrait and voice rights must first be confirmed, and the user must personally upload, create, and select the digital human and voice in Topview or the chosen digital-human platform. The workflow must not upload the original portrait or audio recording. Only after the user confirms that the platform assets are ready may an available adapter be invoked to generate explanatory segments. Then composite the visual explanations, subtitles, and presenter window, and finally verify the aspect ratio, duration, audio, subtitles, and presenter cropping for the target platform. When paid generation is involved, follow the runtime tool's cost-confirmation rules.

When the user wants a finished animated explainer without an on-camera presenter, such as science, history, or brand-story content, follow the animated-explainer production guidelines in : confirm the script and facts, select a voice tone after audition, confirm the visual anchor, and only then generate assets in batches and render. Music may use only locally synthesized audio, tracks from a platform music library, or tracks the user is authorized to use, and attribution requirements must be recorded. After the master version is complete, derive versions for each platform: YouTube at 16:9 with SRT subtitles, and Douyin, WeChat Channels, and Xiaohongshu at 9:16 with burned-in subtitles. Record them in the publishing matrix.

Read quality-gates.md in full and check:

Every failed check must include a corrective action and the result of a recheck. After the review passes, wait for final-draft confirmation.

Invoke to save milestone deliverables and verify them by reading them back. Report only files that have been confirmed to exist and contain non-empty content.

Read publishing-and-review.md in full. For WeChat Official Account drafts, after the user authorizes the operation, preferably invoke to format the content and write it to the draft box. If the module or its dependencies are unavailable, deliver a manual publishing package. For finished videos, invoke : first deliver platform-specific publishing packages, and after the user authorizes each platform, fill in the metadata, upload privately, or schedule publication according to the granted authorization level. Do not mass-send on any platform.

After publication, invoke . Record the raw data and compare content on the same platform, with the same content type, and from a similar period. Output key conclusions, attribution, comparisons, next actions, and hypotheses to validate.

Read style-learning.md. A one-time manual edit is recorded only as an observation; it can be promoted to a stable preference only if it recurs or the user explicitly confirms it.

At every pause or end, report the current task, completed stages, artifact paths, known fallbacks, and the single next action. Use the following statuses: