---
name: render-cuebook-market-post
description: Render or freeze original Frame body copy. Default to one selected FramePreviewV1 and preserve its exact public copy during release freeze; generate three meaning-locked PostV1 siblings only when the creator explicitly requested alternatives. Preserve the creator's viewpoint and keep evidence, policy, settlement, and assistance provenance backstage. The only public destination is Frame. Do not use for social-platform variants, research without writing, living-author imitation, automated publishing, or invented market facts.
license: Proprietary. Cuebook internal; see the repository README for terms.
---

# Render Cuebook Market Post

Turn a supported market event into a readable point of view. Evidence decides what can be said. The event and market tension decide the shape. A profile can tune selection and rhythm; it cannot supply facts or authorize imitation.

## Workflow

1. Resolve the assigned Frame output from ContentRecipeV1. The public plugin path accepts only `frame/publish_candidate_set`; do not generate or mention social-platform variants. Apply language, flavor, and selected profile/visual refs without changing claim strength. Preserve recipe, program, content-item, opportunity, and input-artifact refs in `lineage`.
2. Run the cue through `validate-cuebook-projection`. A `reject` produces no public draft. A `caution` can produce only a watch, debate, or clearly conditional draft.
3. Run `route-cuebook-narrative` and preserve the complete RouteV1 as source triage. `abstain: true` produces no public draft. RouteV1 never owns the final rhetoric or visual grammar.
4. If `ResearchPackV1` is supplied, validate it with `$build-market-research-pack`, preserve its source and fact IDs, and copy `quality_report.decision` into `research_decision`. If the request needs decision-grade comparators, valuation, positioning, or liquidity that the cue does not contain, build a research pack before drafting.
5. Compile or validate `MarketViewSemanticsV1`, then plan the expression with `$plan-cuebook-creator-expression`. This neutral layer locks claim type, subject, comparator, direction, horizon, mechanism, evidence breadth, speaker ownership, action tense, and settlement eligibility before prose changes shape.
6. When `TradingThesisV1` is supplied, validate it first and store `THESIS_id@revision` plus its canonical hash in `lineage.input_artifact_refs`. Treat its claim, direction, probability, horizon, invalidation, resolution criteria, and disclosures as locked. A renderer may shorten them; it cannot improve or rewrite them.
7. Store `CEXP_plan_id@revision` and `meaning_fingerprint.fingerprint_sha256` in `lineage.expression_binding`. Follow its ordered narrative primitives, `VoiceSpec`, text blueprint, authorship mode, required caveats, and source-style firewall. If a downstream draft changes the fingerprinted claim, direction, horizon, trade intent, settlement intent, or ownership, block it.
8. Build or import the fact ledger before drafting. Give every usable fact an ID, evidence class, source URL, event time, observed time, freshness state, and allowed wording.
9. Resolve authorship mode from the expression plan. In `cuebook_assisted`, preserve `creator_seed`, Cue and evidence source refs, each proposed `evidence`, `connection`, `countercase`, or `rule`, creator acceptance or rejection, `idea_delta`, and final judgment in internal `assisted_discovery`; force `public_attribution: false`. Only adopted additions enter first-person meaning. If no real creator seed exists, use generated or source-transformation disclosure; never invent first-person discovery.
10. Fetch only the small amount of live context still required by the expression plan. Timestamp every price, probability, spread, flow, or news update. Preserve whether a displayed current price is a live trade, last close, midpoint, or another quote type.
11. Draft the Frame body from the selected blueprint. Treat the creator's explicit viewpoint as the thesis, not as a claim to disprove before writing. Use adopted Cue connections and Cuebook evidence to strengthen, connect, narrow, or qualify it. A mechanism or scenario may remain the creator's clearly framed inference; do not demand that every reasoning link be established as fact. Only a material factual contradiction may stop or change the view, and then show the conflict to the creator instead of silently replacing their judgment. Make the mechanism and next observable legible without forcing every reasoning job into prose; the paired image owns the visual progression, observed trend, and timing marker. Remove all research-process and assistance narration from the body.
12. Check current Frame policy and record position, commercial relationship, identity, and AI-assistance disclosure states in structured metadata. Never turn those records, source counts, tags, settlement objects, or workflow state into sibling visible sections beside the Frame body.
13. Run four focused passes: semantic lock, evidence, creator lift, and Frame fit. Then score with `evals/rubric.md` and repair every hard failure.
14. When returning structured output, run `scripts/validate_post_artifact.mjs` and resolve all errors. Warnings require a deliberate review.
15. When the user accepts a settleable horizon or threshold, route the finalized prose to `$compile-cuebook-settlement-claim` only when the expression plan says settlement is eligible. Keep the contract as internal Frame metadata. Mention a deadline or level in the body only when it improves the argument; never display a separate settlement panel in the creative result.
16. When the user asks to stage or publish, return the PostV1 to the Frame creation workflow. This skill performs no account operation.

Read `references/rendering-method.md` for evidence classes, angle selection, profile use, and Frame composition. Use `references/skill-matrix.md` as a market-situation reference, not as a substitute for source-first routing.

## Frame Copy Contract

- The final visible Frame has exactly three components: one title, one body, and one paired image. This renderer owns the body; the candidate assembler derives the title and binds the image.
- In Chinese, normally use 260–700 visible characters across three to five short paragraphs (roughly 120–300 English words). Finish sooner when the view is genuinely simple; do not pad, but do not collapse a real causal chain into a caption.
- The body carries the strongest observation, creator-owned viewpoint, causal sequence, horizon, and one confirming or weakening next observable. It does not need to carry all seven reasoning jobs when the image can show the relationship, two to four reasoning beats, and the deadline more clearly.
- The first paragraph is a self-contained Feed lead. Later paragraphs carry the deeper mechanism and future check for the detail surface; do not move that extra density into the image.
- The body and image divide labor. Do not repeat the title, reproduce every image label in prose, or paste a settlement form beneath the argument.
- Keep viewpoint, inference, and observation honest: the viewpoint belongs to the creator; a mechanism, analogy, or future scenario may remain a hypothesis; another Cue remains an external viewpoint until adopted; hard market facts require source-linked evidence.
- Never pad with generic context, repeated warnings, engagement bait, platform language, or workflow narration.

## Selected Preview Freeze

When the input is a confirmed `FramePreviewV1` candidate, do not draft again. Materialize one canonical `PostV1` whose Frame copy reconstructs the exact selected title/body pair, preserve its query/evidence refs, and run release checks only on that copy. Any editorial improvement requires returning a visibly revised preview to the creator for confirmation.

## Autonomous Candidate Mode

When called for explicit alternatives by `$assemble-cuebook-publish-candidates`, use one batch generation pass to render exactly three passed, Frame-ready `PostV1` siblings from the same expression fingerprint and fact ledger. Use distinct openings: conviction first, evidence first, and catalyst/condition first. Do not enter this mode for the default one-preview path.

- Return finished copy only. Do not expose outlines, rejected tensions, repair notes, or requests for another writing round.
- Keep every canonical `PostV1` Frame-sized. The candidate assembler derives a title and its internal body/close split, then exposes only the canonical `title`, combined `body`, and paired `image` projection. Tags and split fields remain internal.
- Resolve routine volatile context through the research pack, Cuebook data, official primary sources, and approved live providers before writing. Repair or omit unsupported facts internally.
- Keep one judgment and one argument spine, then make the forced actor, transmission, market consequence, horizon, and material caveat understandable. Remove repeated setup before removing a reasoning link.
- Run semantic-lock, evidence, human-language, and compactness passes on every sibling. Regenerate a failed sibling; never send a weak option downstream merely to reach three.
- Preserve the same asset, direction, horizon, caveat, authorship, and settlement eligibility across all candidates.
- Run semantic-lock and evidence checks once across the batch, then score voice and compression per sibling. Regenerate only the sibling that fails.

## Hard Gates

- Source ticker or entity mismatch: reject until repaired or mapped to a closer asset.
- Proxy cue without an explicit target, repricing actor, and causal chain: reject.
- Narrow company or sector event forced into a broad ETF: reject or reroute.
- RouteV1 abstention or unknown event classification: block public drafting until the route is repaired.
- Missing or changed expression fingerprint, reassigned speaker ownership, anecdote upgraded into a cohort claim, or an observation upgraded into a trade: block until restored to the locked expression plan.
- Analyst target without a model reason: caution; do not turn it into a thesis.
- Unchanged guidance without a consensus comparison: caution.
- Beat/miss, revision, or reaction language without a comparable period, basis, source, and timestamp: remove or make conditional.
- Trade conclusion without a supported asset expression, horizon, and sufficient liquidity context: keep it as a watch, not an instruction. Invalidation may remain in the structured thesis or settlement artifact without appearing in body copy.
- Frozen thesis derivative with missing thesis ref/hash or changed claim, direction, probability, invalidation, or resolution criteria: block until restored to the canonical declaration.
- Current fact without source and `as_of`: remove it or mark it unavailable.
- Ready market commentary with unknown position or commercial relationship, or a ready policy snapshot older than 30 days: keep conditional until resolved.
- Material number absent from the fact ledger: remove it.
- A reasoning-complete post reduced to selector copy, or a post that omits the forced actor or transmission solely to satisfy a preview budget: reject and restore the full argument behind `post_ref`.
- A settlement footer with an invented or unconfirmed deadline, threshold, session, benchmark, or source: keep it separate in `needs_confirmation`; do not attach it as a ready claim.
- A public draft that narrates Cuebook assistance, idea completion, workflow steps, or accepted/rejected additions: remove that narration. Preserve it only in internal provenance fields.
- A draft or handoff that names X, Xiaohongshu, Reddit, Telegram, a thread, a caption, or any destination other than Frame: reject and restore the Frame-only shape.
- A creative result that exposes labels, tags, evidence ledgers, settlement panels, source counts, quality scores, or workflow state beside the title, body, and image: reject the presentation projection.
- A body that treats the creator's viewpoint as a mistake to correct without a material evidence contradiction: reject and restore creator ownership.
- A post whose prose says wait, watch, avoid, or exit while its settlement claim says immediate long or short: block the pair until action state and claim agree.
- Direct imitation, exact catchphrases, or identity performance for a living author: refuse that part and render original analysis.

## Writing Rules

- Start with a concrete change, judgment, or tension. Avoid an abstract scene setter.
- Optimize the creator's intended judgment before adding caveats. The experience should feel like their idea became clearer, sharper, and better expressed—not like a referee graded it.
- Put one thought in each paragraph. Let the source fact and the market read occupy separate sentences.
- Name the actor who may need to revise, hedge, chase, cut, or wait.
- End on the horizon, next data point, catalyst, or market condition. Do not force an invalidation paragraph or generic warning.
- In Cuebook-assisted mode, keep the creator as the decision owner and the assistance record internal. The public draft must not say Cuebook supplied, inspired, completed, strengthened, or corrected the idea.
- Keep the language conversational without pretending to have private access or personal experience.
- Build the public argument as `claim -> proof -> market consequence -> next condition`. Do not add marketing CTAs or engagement bait unless explicitly requested.
- Make the market movement concrete: name the participant who is likely to hedge, reprice, rotate, chase, cut, or wait; name what exposure or balance-sheet constraint changes; then connect that action to the selected asset. Do not jump from news to direction in one sentence.
- Let the expression plan select the actual rhetorical order. `claim -> proof -> consequence -> condition` is a fallback, not a template to repeat across every post.
- Do not use `我认错`, `哪里认错`, `错了怎么办`, or equivalent self-correction headings. If a selected format explicitly requires a risk boundary, phrase it neutrally as a material condition and keep it separate from the creator's main argument.
- Avoid stock AI phrases such as `值得关注的是`, `从机制上看`, `核心逻辑在于`, `传导路径`, `验证路径`, and repeated `不是 A，而是 B` framing.
- Do not add a ticker, target, direction, motive, or hard number merely to make the post sound tradable.

## Output Contract

Return `PostV1`:

```json
{
  "schema_version": "post-v1",
  "lineage": {"artifact_id": "POST_...", "program_ref": null, "content_item_ref": null, "opportunity_refs": [], "input_artifact_refs": [], "thesis_binding": null, "expression_binding": {"plan_ref": "CEXP_...@r1", "fingerprint_sha256": "sha256:..."}},
  "brief": {"platforms": ["frame"], "content_class": "market_commentary", "temporal_mode": "realtime", "language": "zh-CN", "as_of": "", "reader": "", "decision_use": "", "research_pack_ref": null},
  "gate": {"decision": "pass | caution | reject", "checks": [], "repairs": []},
  "research_decision": "ready | conditional | blocked | null",
  "policy_gate": {"decision": "ready | conditional | blocked", "checked_at": "", "rules_checked": [], "repairs": []},
  "disclosure_state": {"position_status": "unknown", "position_text": null, "commercial_status": "unknown", "commercial_text": null, "identity_status": "unknown", "ai_assistance_status": "unknown", "public_disclosures": []},
  "route": {"schema_version": "route-v1", "taxonomy_version": "market-narrative-v2", "cue_id": "", "event_type": "", "event_confidence": 0, "candidates": [], "reasoning_lenses": [], "render_shape": "", "required_context": [], "hard_numbers": [], "abstain": false, "abstain_reason": ""},
  "fact_ledger": [
    {"id": "F1", "claim": "", "evidence_class": "source | verified-live | derived | hypothesis", "source_url": "", "as_of": "", "freshness": "current | stale | unknown"}
  ],
  "assisted_discovery": {"mode": "none | cuebook_assisted", "creator_seed": null, "cuebook_contribution": null, "creator_judgment": null, "idea_delta": null, "final_trade_idea": null, "fact_refs": [], "public_attribution": false},
  "angle": {"tension": "", "forced_actor": "", "why_selected": "", "profile_rule_ids": []},
  "drafts": {"frame": ""},
  "draft_evidence": {"frame": []},
  "watch_items": [],
  "quality_report": {"scores": {}, "hard_failures": [], "revisions": []},
  "publication_state": "ready | conditional | blocked"
}
```

Set `publication_state` to the stricter result from `gate.decision`, RouteV1 abstention, `research_decision`, and `policy_gate`, then apply disclosure guards. For a blocked result, leave all drafts empty and explain the repair. For a conditional result, use conditional language.

## Resources

- `templates/brief-template.md`: normalized request and source boundary.
- `references/rendering-method.md`: evidence, angle, profile, and Frame method.
- `references/post-v1.schema.json`: structured output contract.
- `references/skill-matrix.md`: market-situation examples and context needs.
- `scripts/validate_post_artifact.mjs`: deterministic PostV1 invariant checks.
- `tests/validate_post_artifact.test.mjs`: regression tests.
- `evals/trigger_cases.json`: positive, negative, neighboring, and adversarial routing cases.
- `evals/expected_artifacts.json`: PostV1 contract expectations.
- `evals/rubric.md`: content quality gate.
- `evals/failure_cases.md`: stable regression cases from Cuebook.