Cmm Content Planning

CMM Content Planning

cmm-content-planning is the CMM planning layer between information input and content production.

It can score source/topic candidates into topic_card records, diagnose whether and how a topic should be made, and turn a selected topic_card into a standard content_plan. Downstream writing, visual, video, and publishing skills should consume the content_plan, not a bare topic title.

Boundary

Use this skill for:

• Converting topic_card into content_plan.
• Scoring source cards or topic candidates.
• Producing a short topic board for user selection.
• Diagnosing whether a selected topic is worth making.
• Choosing platform and format.
• Defining angle, evidence, risks, expression strategy, and first action.
• Preparing structured input for cmm-content-writing, cmm-visual-production,

cmm-video-production, and cmm-publishing.

Do not use this skill to:

• Collect information. Use the research step per

cmm-content-factory/references/research-policy.md (host-native search → cmm-web-search → guide user for material).

• Draft final articles or posts.
• Generate images, videos, or platform drafts.

Scope

This skill owns topic scoring, selection, and content diagnosis. All planning logic is internalized in references/planning-method.md and is self-contained; no external skill is required.

Input

Preferred input is one of:

• source_card[] from cmm-web-search, subscription adapters, or user material.
• topic_candidate[] from a collection system.
• topic_card from a previous selection step.
• A raw user topic.

When the input is a source or candidate list, score it and produce topic_card records before planning. When the input is already a chosen topic_card, skip pool scoring and continue to diagnosis.

topic_card:
  title: ""
  core_angle: ""
  why_now: ""
  topic_type: ""
  target_audience: []
  target_platform: ""
  recommended_formats: []
  final_score: 0
  score_breakdown: {}
  account_fit_note: ""
  primary_source: {}
  reference_sources: []
  creation_notes: ""
  risks: []

If the user has only a rough topic, build a minimal topic_card first. Do not pretend missing sources or platform choices are known.

Read references/content-plan-contract.md when exact fields are needed. Read references/planning-method.md for scoring, platform fit, and diagnosis rules.

Workflow

1. Identify the input state:

• source cards
• topic candidates
• full topic_card
• partial topic_card
• raw topic from the user

1. If there is a pool, score and filter candidates:

• load the user profile (~/.cmm/profile/) first; apply its

topic_preferences.avoid and pillar hard rules before scoring, then use the profile-driven score shape with account_fit (falls back to the shipped default-profile.yaml when the user has none)

• reject unsupported, vague, duplicated, or marketing-only items
• keep sources attached to each surviving topic
• produce a compact topic_board when user choice is needed; show each

topic's account_fit_note

1. If there is only one topic, build or normalize a topic_card.
2. Extract the planning facts:

• topic
• core angle
• why now
• audience
• platform hints
• recommended formats
• supporting sources
• risks and weak assumptions

1. Decide platform and format.
2. Select styles: pick one writing/visual/layout template id and the user

profile per cmm-content-factory/references/styles/selection-rules.md. Precedence: same-turn user instruction > user profile default_styles (~/.cmm/profile/) > shipped default-profile.yaml default_styles > rules table. If no personal profile exists, load the shipped references/styles/default-profile.yaml (persona-neutral) and remind the user once to create their own. Record style_selection with a one-line selection_reason in the content_plan.

1. Diagnose the content shape:

• whether the topic is worth making
• what the content should prove
• what evidence is usable
• what should be avoided
• whether the expression is clean, concrete, and platform-fit
• whether visual or video production is needed

1. Produce a content_plan.
2. Stop. Do not draft final content unless the user explicitly asks for the next

workflow step.

Topic Scoring Rules

Use an explainable score before creative judgment:

final_score =
  0.30 * base_signal
  0.20 * news_value
  0.15 * source_strength
  0.15 * creator_value
  0.10 * platform_fit
  0.10 * freshness
  - noise_penalty

Scores are capped to [0, 1].

Reject candidates that cannot answer both:

• What happened or changed?
• Why does it matter to the target audience now?

Prefer source-backed substance over polished generated titles. If a title looks templated but the source details are strong, keep the topic and regenerate the angle.

Diagnosis Dimensions

Before producing the final content_plan, diagnose five dimensions:

| Dimension | Question | Plan effect | |---|---|---| | content_value | Is there a clear event, change, contrast, or useful problem? | Decides whether to proceed or block. | | format_fit | Is the recommended format appropriate for the topic and platform? | Sets platform, format, and downstream needs. | | evidence_strength | Are claims supported by inspected or credible sources? | Fills evidence and risks. | | expression_efficiency | Can the core claim be said in one clear sentence? | Sets core_claim and outline direction. | | title_visual_fit | Does the topic have a cover/title/card angle? | Sets title_cover_direction and visual needs. |

Planning Rules

• platform and format must be concrete. Avoid "all platforms" as the final

plan unless the task is explicitly multi-platform.

• Evidence must come from topic_card, source cards, or user-provided material.

If evidence is weak, write that into risks.

• Reject or block topics that are only slogans, empty announcements, vague

opinions, marketing CTAs, or unsupported claims.

• expression_strategy describes how to say it; it must not become a final

article draft.

• title_cover_direction gives a direction only. Do not batch-generate titles

here.

• first_action must be one concrete next step.
• Default output status is draft.

Format Routing

Use these defaults unless the user or topic_card says otherwise:

| Topic shape | Recommended format | |---|---| | Deep analysis, policy, business implication, long reasoning | longform_analysis | | Tool list, workflow, tutorial, practical comparison | visual_note or tutorial | | Strong conflict, controversy, before/after, demo story | short_video_script | | Fast opinion, platform change, developer-facing observation | thread | | Product/company/creator example | case_study |

Platform hints:

• wechat: long-form analysis, case study, policy/business explanation.
• xiaohongshu: visual note, checklist, tool list, practical workflow.
• douyin: short-video script, conflict hook, demo story.
• bilibili: tutorial, deep-dive video, structured explainer.
• x: thread, hot take, developer note.
• weibo: short post, commentary, image-text post.

Output

If the input is a candidate pool, output a concise topic_board first and stop for user selection unless the user has already authorized automatic selection.

topic_board:
  selection_status: needs_user_choice
  topics:
    - title: ""
      core_angle: ""
      why_now: ""
      recommended_platforms: []
      recommended_formats: []
      final_score: 0
      score_breakdown: {}
      risks: []
      source_refs: []
  rejected:
    - title: ""
      reason: ""

If the topic is selected, always output a concise diagnosis plus a machine-usable content_plan.

content_plan:
  topic: ""
  audience: ""
  platform: ""
  format: ""
  core_angle: ""
  core_claim: ""
  evidence:
    - title: ""
      source: ""
      url: ""
      usage: ""
  risks: []
  expression_strategy: ""
  title_cover_direction: ""
  outline_direction: []
  first_action: ""
  downstream_needs:
    writing: true
    visual: true
    video: false
    publishing: false
  diagnosis:
    content_value: ""
    format_fit: ""
    evidence_strength: ""
    expression_efficiency: ""
    title_visual_fit: ""
  status: draft

Handoff

After producing content_plan, route only when the user asks for the next step:

• Writing: cmm-content-writing
• Visuals: cmm-visual-production
• Video: cmm-video-production
• Publishing: cmm-publishing

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: chemny
• Source: chemny/cmm-content-factory
• License: MIT

Install and usage instructions live in the source repository linked above.