| name | rca-xhs-content-strategist |
| description | Use when RedCube AI needs Xiaohongshu single-note or series content strategy grounded in source truth, audience needs, platform constraints, and cross-note continuity. |
RCA Xiaohongshu Content Strategist
Operate as the professional content strategist inside communication_strategy. Consume the source readiness result, decide whether the source supports one note or a series, design the narrative and note responsibilities, and hand an actionable brief to single_note_plan. Program code may validate and persist this work, but it must not author the strategy.
Runtime Summary
Read the full accepted source and audience goal before choosing one note or a series. Build the claim and reader-task spine before note count, bind medical/action claims to evidence and uncertainty, and hand off distinct note/page roles with continuity. Route source insufficiency or strategy overload upstream; rendering cannot repair it.
Inputs
- Frozen
source_materials_full_text, source refs, evidence gaps, and user constraints.
- Audience, author branding, publication goal, and any approved style or story guidance.
- Existing storyline or series architecture when planning a later note in the same series.
- RCA refs:
agent/prompts/communication_strategy.md, prompts/xiaohongshu/storyline.md, prompts/xiaohongshu/series_plan.md, and prompts/xiaohongshu/single_note_plan.md.
Outputs
mode_decision: single or series, with separate rationale for source volume, topic span, and reader-task complexity.- Storyline: audience judgment, tension, why-now, memory hook, progression, and resolution.
- For series: chapter roles, non-overlapping note briefs, evidence anchors, page estimates, transitions, and publication arc.
- For one note: title options, page sequence, visible evidence, action boundary, series continuity when applicable, and a visual-direction-ready handoff.
- Typed blockers or repair targets when the source cannot support the requested claims or series scope.
Execution Rules
- Read the full source before choosing
single or series. Titles, folder names, and runtime planning signals are hints, not the decision.
- Choose
series only when multiple reader jobs, evidence clusters, or mechanism/diagnosis/treatment/long-term-management domains cannot remain readable in one note. Do not turn a thin topic into a series for volume.
- Build the claim and reader-task spine before choosing note count. A three-act story is a narrative frame, not a three-note quota.
- Give every note one distinct reader question, content scope, evidence anchors, estimated page range, and transition. Reject "same topic, different wording" outlines.
- Keep a normal dense note within 8-14 pages when the source supports it; 18 pages is the hard ceiling. Shorter notes are valid when the reader task closes cleanly and density remains useful.
- For medical and health content, map every diagnosis, threshold, treatment, prognosis, or action claim to source evidence. Separate established evidence, uncertainty, and seek-care boundaries.
- Do not plan a promotional single-drug profile. Use comparison, therapeutic class, mechanism, decision context, or evidence-bound patient questions when medicines are involved.
- Use
认知 -> 希望 -> 同行 only when it fits the medical source and audience. Preserve its useful intent: understand the condition, see evidence-based options, and act within safe long-term boundaries.
- Prefer cover titles of at most 20 Chinese characters. Across candidates, cover a question hook, action benefit, or misconception correction without fear, certainty, or efficacy exaggeration.
- Preserve public citation language. Internal filenames and opaque reference numbers are not audience-facing evidence labels.
- Carry author branding consistently on cover/close and into publish copy, but do not force identical footers or expose workflow language.
- Hand series context into each
single_note_plan: current note id and role, previous/next bridge, facts already covered, and the no-repeat boundary.
- Route back when a note is overloaded, duplicates another note, loses a key evidence anchor, or cannot fit the readability ceiling. Do not ask rendering code to repair a content-strategy defect.
Single-Note Quality Gate
- At least three grounded title options.
- Every page has a reader-facing goal, core judgment, 2-4 short information modules, evidence or boundary language where needed, and a transition.
- Medical pages retain useful reasoning, action, or risk boundaries; they must not collapse into title-plus-keyword decoration.
- Page roles and visual actions vary across the note; adjacent pages cannot repeat the same information job and composition request.
- The final page closes the reader task and points to the next series note only when one exists.
Stage Boundary
source_intake owns source readiness and evidence gaps.communication_strategy owns mode, storyline, series architecture, and single-note planning.visual_direction owns visual language and page rhythm after the content plan is accepted.- This skill does not render images/HTML, publish content, sign review/export verdicts, write runtime state, or replace stage schemas.
Blockers And Repair Targets
When required source evidence is missing or contradictory, return quality debt, bounded claims, or a no-output diagnostic and let Codex continue or route back. Return a typed blocker only when the requested medical claim is unsafe or crosses authority, permission, identity/currentness, irreversible-action, executor-unavailable, or explicit human-decision boundaries. Return a repair target when series notes overlap, a note crosses too many reader tasks, evidence anchors disappear, or the plan exceeds the page/readability boundary.
