The storyboard step of the comic-author suite: take a LOCKED, user-approved\noutline_spec and produce the artifact that IS the authoring source of\ntruth — a fixed page order authored as its own section BEFORE any per-panel prose, the MOTIF STATE\nTABLE (the master per-panel ledger that pins every continuity variable for all panels before any bake),\na fixed 9-field per-panel spec, and one deduped canonical asset contract. This is the deep\nlayer the thin comic-author orchestrator SOP defers to; it ports the rigor of aris_movie's\nmovie-storyboard-creator (抽卡 lineages → score → gate) and -v4 (locked-inputs precondition → beat→panel\nbudget → deterministic spec) onto stills, and it emits the panel_spec fields the downstream\ncomic-blueprint-author and comic-director\nconsume. The downstream baking spiral is not this skill's job — this is pure authoring + ledger.
\n\nTwo-stage role (the N1 contract): this layer's gate runs TWICE. Right after the outline locks, this\nskill authors the storyboard and runs
\n--gate storyboardas a PROVISIONAL structural pass — the\nstoryboard MAY reference draft (declared-but-unlocked) assets;panel_assets_referenceableis\nunscorable, is leftnull, and the fuser SKIPs it (nolockedflip yet). Then, after the asset layer has\nlocked everything and OUTLINE_FINAL_LOCK has re-confirmed the outline, the FINAL asset-resolution\nvalidation re-runs--gate storyboardwith all four dims scorable and flips the storyboard tolocked.\nThe hard locked-asset barrier stays downstream, before blueprint authoring\n(comic-authorsteps 6a–7).
\n\nCardinal lesson baked in as a gate, not prose: decompose beats→pages→panels and 抽卡 K orderings by\nfeasibility, then fill ONE panel at a time — never batch-author-then-look. The MOTIF STATE TABLE is the\nconcrete guard: it pre-commits a per-dimension-per-panel ground truth (mug heat / executor bounce /\nTok|yo evidence / DDL spine / exact_parse / claim_delta) with machine-checkable global invariants\n(DDL monotonic never-rewind; bounce S02 = the film's ONLY MAX; REJECT↔ACCEPT mirror), so the audit reads\nintent off a table instead of inferring it.
\n
locked+approved outline_spec ─▶ ① PRECONDITION (refuse if outline unlocked / unapproved; draft assets OK here)\n ▼\n ② DECOMPOSE beats → pages → panels (beat→panel budget arithmetic)\n ▼\n ③ 抽卡 K page/panel-ORDER lineages (Codex xhigh) → score(feasibility/fidelity/novelty) → keep\n ▼\n ④ PAGE ORDER FIRST (a fixed section, page-type + L/R world per page, BEFORE prose)\n ▼\n ⑤ MOTIF STATE TABLE (master ledger; monotonic DDL; bounce-uniqueness; mirror locks)\n ▼\n ⑥ PER-PANEL 9-FIELD SPEC (one unit at a time; DONE panels locked; baked-vs-gate contract)\n ▼\n ⑦ CONSOLIDATE ASSET_REQUESTS (one canonical owner per token; naming 铁律)\n ▼\n ⑧ EMIT storyboard_spec + motif_ledger + N panel_spec (+ continuity_constraint)\n ▼\n ⑨ GATE ×2: PROVISIONAL structural pass now → (assets lock + OUTLINE_FINAL_LOCK)\n → FINAL asset-resolution pass flips `locked`\n抽卡 K (lineages of page/panel ordering) — K_LITE=3, K_BALANCED=6 (default), K_MAX=8, K_BEAST=12\n(ported verbatim from movie-storyboard-creator). Effort tier governs count of orderings, not depth.
FEASIBILITY_FLOOR = 5 — a lineage scoring feasibility < 5/10 is dropped (e.g. it needs arbitrary\nun-gateable text rendering, exact UI replication, or face identity it can't anchor). If <2 survive,\nre-prompt once on the same Codex thread; still <2 → tell the user to lower --effort or relax the\nbanlist. Never fall back to a Claude-generated lineage (violates reviewer independence).
REVIEWER = the Codex CLI at model_reasoning_effort: xhigh for the 抽卡 brainstorm + the cross-layer\ngate — it pins NO model (it follows the local codex config, currently gpt-5.6-sol); Gemini auto-gemini-3\nwhere a second family is needed. Never downgrade the effort tier (reviewer-routing).
PAGE_TYPES (authoring vocabulary) = {cover, single, 2-up, vertical-3, grid2x2, big-frame, endcard}.\nThese are NOT the compiled IR's page types — schemas/comic.schema.json pages[].type enum is\n{cover, single, grid, grid2x2, feature, finale} — so the compiler MUST project them (an unmapped token\nFAILS jsonschema):
| storyboard token | \ncompiles to pages[].type | \n
|---|---|
cover / single (incl. full-bleed) / grid2x2 | \nsame name | \n
2-up / vertical-3 | \ngrid | \n
big-frame | \nfeature | \n
endcard | \nthe PRIOR page's closing.image — the shipped-reference convention: comic_m3_audit compiles S22's endcard into P_B12.closing.image (18 pages ship, not a 19th endcard page). finale stays a legal enum value for a standalone end page, but closing.image is THE convention to copy. | \n
WORLDS = {warm, dark-cyber, seam, starfield} (human = warm, ARIS = dark-cyber, explicit seam at\nhand-off/payoff, starfield for the wiki/finale) — the two-world palette grammar inherited from the outline.
MOTIF_TABLE_COLUMNS (per panel, verbatim) = Panel | <motif-A> | DDL | <motif-B> | exact_parse | claim_delta | bounce\n— for the worked example: Panel | Mug | DDL | Tok|yo | exact_parse | claim_delta | bounce. The two metric\ncolumns (exact_parse, claim_delta) never co-mingle (different metrics, different value sets).
PANEL_TEMPLATE = the fixed 9 fields (+ title anchor): text_mode / content_blueprint /\nexpected_literals / scene / characters / bubbles{zh,en} / side_narration{zh,en,trait} / motifs /\nasset_requests. No free prose outside these fields. (cover/finale panels add safe_zones/endcard fields.)
OUTPUT = story/STORYBOARD_DRAFT.md (the human artifact) + wiki nodes (storyboard:<slug>, motif:<slug>,\npanel:<slug> × N, optional cont:<slug>). Versioned per output-versioning.
This skill is pure authoring + ledger, and it operates ONLY on converged upstream (convergence is the\nupstream gates' job, per acceptance-gate):
outline_spec is not status: \"locked\", OR it lacks\nthe explicit user-approval stamp (the user-first gate — never proceed without a recorded\n✅ APPROVED by user <date>). Whitelisted assets may still be draft/declared at the PROVISIONAL pass —\nthat is the N1 two-stage contract; an unlocked asset only blocks the FINAL asset-resolution validation\n(and everything downstream of it).character_asset_ids / scene_asset_ids / prop_asset_ids plus assets you newly request as\nasset_requests. Never free-describe an asset inline. If a panel needs an asset that doesn't exist yet,\nadd it to asset_requests (which reopens the asset layer) — do not invent it silently.narrative_beats[], beat_table (the 6-column table: Beat / Panels / layout /\nworld / content+台词 gist / ARIS-feature-as-story-cost), motif_arcs (the per-shot continuity values),\ndesign_constraint (the dual-identity), title / logline / tagline.Load the outline_spec. Verify status == \"locked\" AND the user-approval stamp is present. (Whitelisted\nassets may still be draft — locked is the FINAL pass's demand, not an authoring precondition; N1 two-stage.)\nOn any failure → HALT with the precise offending node id so the orchestrator routes\nback to the outline layer. Record provenance in the header: source outline id, the prior\ncross-model verdict it survived (real example: codex critique returned REVISE; \"全部 fix 落地\"), and the\n✅ APPROVED by user <date> line.
For each outline beat, read its layout token to set its page type and per-page panel count, then assign\nshot/panel ids. Compute the page budget: a beat tagged vertical-3 owns 3 panels on one page; 2-up owns 2;\ngrid2x2 owns a recap page; single/full-bleed/endcard own 1. The centerpiece beat may expand\n(real: B08 went 4 panels → 5 single-read pages, adding a NEW S12a audit-entry panel, with the 2×2 kept\nONLY as a recap/hero page). Clamp the global total to the project's tier window; if oversize, collapse the\nlongest multi-panel beat first; if undersize, split the beat with the most slack. Every panel must land on\nexactly one page with a stable page id (P00_cover, P01_b02, …, P03_end).
Pass file paths only (the locked outline, the style bible, the wiki banlist of prior failed orderings) to\nthe Codex CLI at xhigh (no model pin — local codex config) — reviewer independence: Codex forms its own\nreading, never gets your summary. Demand\nK (= the tier constant) alternative page/panel orderings of the same beat set, each differing on ≥2 of:\npage-rhythm axis (cover/2-up/grid/vertical-3/big-frame mix) · cast-presence axis (which panels show the\nduo vs solo) · world-alternation axis (warm/dark/seam sequencing) · reading-order axis. Score each lineage\n0–10 on Feasibility (lower if it needs un-gateable arbitrary text / exact UI replication / face identity),\nBrief-fidelity (does it preserve the locked beat boundaries + the dual-identity), Novelty-vs-banlist\n(penalize ≥0.6 Jaccard overlap with a prior failed ordering). Drop any lineage with feasibility < 5.\nPromote the top survivor to the active page order; keep the rest as candidate notes (failed orderings become\nbanlist memory so the next run never re-proposes them). The 抽卡 here is over orderings, not panel content —\npanel content is filled deterministically in ⑥.
Author the chosen ordering as a separate, fixed section that comes BEFORE per-panel prose (the\npage_order field of the storyboard_spec). Assign each panel to its stable page id + page TYPE + per-page\nL/R world. End with an explicit TOTALS line: panel count · page count · how many NEW bakes vs reused DONE\nimages. This section is the authority the compiler and the gate check the eventual comic.json against.
Author the MOTIF STATE TABLE as the single source of truth for the continuity dimensions across all\npanels, before writing any per-panel prose. One row per panel; the columns are MOTIF_TABLE_COLUMNS.\nEncode every machine-checkable invariant directly in the table + its conventions block:
Mug): track ONLY the continuity-bearing\ninstance (the ML RESEARCH motif mug: hot two steam wisps = temperature origin → fading one thin wisp\n→ cold no steam + milk skin, a B01↔S20 bookend = \"the hardest continuity constraint\"). Disambiguate\nmotif-vs-env: any look-alike that is just an environment prop (the </> mug) is tagged env and is NOT\ntracked — this is the absence ≠ drift / design-aware gate rule made concrete (a naive holistic gate\nwould false-flag it).T-24:00:00 → 19:42 → 19:42 → 18:55 → 18:40 → 18:30 → 17:46 → 17:02 → 16:21 → 16:05 → 05:12 → 02:58 → 02:41 → 01:57 → 00:42 → 00:27(submitted)). Two consecutive equal\nvalues are allowed (a \"same-instant rhyme\"); an increase is a hard veto. If any draft value would rewind,\nfix it inline and annotate the fix (real: S06 DDL T-15:27 → T-18:40 purely so 18:40 > S07's 18:30).exact_parse carries {0.60, 0.71, 0.66, 0.78, 0.89};\nclaim_delta carries {+6.2, +1.4}. They are different metrics; writing one into the other's column is a\nveto. (Real: S10 shows exact_parse 0.71→0.66 while claim_delta stays —; S14 shows\nclaim_delta +6.2→+1.4 while exact_parse stays —.)bounce = MAX appears exactly\nonce (real: S02 是全片唯一 MAX). Every later high is annotated \"below S02\". After the fall, NO new peak\nis allowed — a post-fall celebration must be written as arrested (real: S12a = attempted celebration, ARRESTED before becoming a jump, the codex fix that removed an earlier \"MAX\" there). The arc closes at the\nsmallest bounce (real: S21 = smallest bounce, the mirror landing of S02).REJECT card ↔ ACCEPT card =\nexact same diagonal stamp geometry/angle (real: S11 ↔ S16, both from the shared verdict-card geometry,\nthe audit hinge); S02 unique-MAX bounce ↔ S21 smallest bounce; a labeled wiki map ↔ its wordless twin\nshare node coordinates programmatically derived from one JSON (real: S16b ↔ S22 from\nwiki_starmap_nodes_v1.json, 禁目测 — no eyeballing).Walk the page order and author each panel in the fixed template — finish one panel fully before starting the\nnext (this is the \"never batch-generate-then-look\" lesson at authoring time). Each panel carries:
\ntext_mode ∈ {baked, html} — baked = the image model renders the in-frame text; html = the\npanel is a pure scene and ALL text lives in the HTML overlay (zero glyphs in the image).content_blueprint — which canonical SVG asset is instantiated + the exact literal layout (e.g.\n\"from assets/ddl_widget_template_v1.svg instantiate T-24:00:00, amber on a navy widget, screen-corner\").expected_literals — the blind-transcribable tokens that will be gated (see the contract below).scene — the staged environment (density target, books/props, the warm/dark world rendered).characters — who is present + their motif-table state (bounce, pose); note who is absent and why.bubbles{zh,en} — speaker + style + bilingual line. Skill names NEVER appear in a bubble or in-frame\n(铁律) — they live only in side_narration.side_narration{zh,en,trait} — first sentence = what happened; second = the ARIS capability bound to\na story cost; the trait tag names the bound skill. (The 侧栏讲解铁律 inherited from the outline.)motifs — restate that panel's row of the MOTIF STATE TABLE (must agree with ⑤ exactly).asset_requests — the canonical asset names this panel consumes (whitelist only).In the human STORYBOARD_DRAFT.md, a DONE / reused panel's row carries ONLY status + side_narration +\nmotifs (as baked) — locked, never retro-edited (real: S12–S15, baked images promoted from a grid to single\npages; their square crops stay in the existing 2×2 recap; no regeneration). A DONE panel that is edited is a\nhard veto. The wiki panel_spec node is NOT reduced: validate_wiki.py PAYLOAD_REQUIRED[\"panel_spec\"]\nunconditionally requires all 13 fields for EVERY panel_spec (no DONE exemption), so a DONE/reused panel's\npanel_spec.payload must still carry every field, holding its prior locked values unchanged (never retro-edited).\nThe reduced 3-field form is a STORYBOARD_DRAFT.md convenience only.
expected_literals holds ONLY the salient, large, blind-transcribable tokens; everything visible-but-not-\ngated is explicitly excluded in writing:
Tok|yo chip is sized ≤1.5% frame height, \"below\nblind-review transcription threshold\", NOT in expected_literals (it is a deliberate foreshadow, no\ncallout, no arrow); it becomes hero/gated only at S10.SUBMITTED; the ✓ is decorative non-ASCII and is out of contract.expected_literals (hard veto).Collect every panel's asset_requests into ONE canonical, deduped, authoritative list. Apply the naming\n铁律: declare all draft-era name variants void and map them to the single canonical name; pin one\nparametric source per recurring motif (real: ONE ddl_widget_template_v1.svg is the sole render source for\nthe entire 18-instance countdown timeline; ONE stamp_family_v1.svg for ALL verdict stamps — no two stamp\nsets, and it must match the already-baked S12–S15 stamps; ONE wiki_starmap_nodes_v1.json is the single truth\nsource shared by the labeled map S16b and its wordless twin S22). Flag P0 blockers (real:\nresearcher_chibi_canonical_ref_v001.png missing — only the duo had a canonical ref) and enumerate\nzero-new-work reuse (DONE panels, the existing recap page, S01's baked image reused as a set-continuity\ncondition ref for S20/S21 same-room/same-desk/same-camera).
Write the wiki nodes (§ below) and append the author edges. The storyboard gate is a score-FUSER: it\nwalks reviews edges into the target, fuses the pre-existing review:* score-nodes, and HARD-FAILS if\nzero are attached (\"no reviews — gate is a score-fuser, not a reviewer\"). So before invoking it:
asset_request (DECLARED), but locked\nis NOT required — panel_assets_referenceable (resolves AND locked) is unscorable, is left null, and\nthe fuser SKIPs it (the provisional verdict rides on the other three structural dims; no locked flip\nyet). The FINAL pass runs only after the asset layer has driven every requested asset to locked AND\nOUTLINE_FINAL_LOCK has re-confirmed the outline (comic-author steps 5–6b):\nnow all four dims are scorable, an un-locked ref hard-vetoes on _unresolved_asset_refs, and an approve\nverdict flips the storyboard to locked.panel_assets_referenceable stays null — see ⑨.0) and persist them as a review node:\nnode_id: review:<slug>, payload {target_node_id: \"storyboard:<slug>\", reviewer, gate_kind: \"storyboard\", review_scores:{panel_assets_referenceable, global_policies_valid, panel_count_band_aligned, continuity_chain_well_formed}}, with a reviews edge review:<slug> → storyboard:<slug>.python3 / call comic-cross-layer-gate\n<storyboard:slug> --gate storyboard to fuse + adjudicate against the EXACT predicate below. On an\napprove verdict (the gate's verdict set is {approve, revise}) at the FINAL pass, the gate flips\nstoryboard_spec.status → locked; a PROVISIONAL-pass approve leaves it under_review (⑨.0), and\nrevise sets/keeps under_review at either pass. (Canon ⑥ FLIP: locked is a STATUS reached by the\nadvance verdict, never a verdict name.) The gate (a different model\nfamily) — never this authoring agent — acquits (artifact-integrity).\nReviewer routing for any model-family fan-out: Claude (one lens) ‖ Codex CLI xhigh (no model pin — local\ncodex config, currently gpt-5.6-sol) ‖ Gemini\nauto-gemini-3, file paths only (reviewer-routing).storyboard) — quote comic-cross-layer-gate --gate storyboardThis is not this skill's rubric to define — comic-cross-layer-gate\nis the SOLE gate authority, and --gate storyboard is STRUCTURAL, CC-only (no visual reviewer; no pixels\nyet). Verdict set {approve, revise}. Under the N1 contract it is invoked TWICE (provisional → final; the\nwhich-pass rules are ⑨.0). The text below is quoted verbatim from that gate so this skill authors\nto the real predicate — if the two ever diverge, the gate wins.
Scored dimensions = exactly FOUR; each scored 0–5; ADVANCE (APPROVE) iff ALL FOUR ≥ 4. These four are\nfile-system facts the gate computes, not reviewer opinion:
panel_assets_referenceable — every asset ref in each panel resolves AND is status: locked.\nScored at the FINAL pass only; at the PROVISIONAL pass it is left null/SKIPped and the declared-check\n(whitelisted or complete asset_request) applies instead — see ⑨.0 for the two-stage ordering.global_policies_valid — global_policies fields match expected (text-mode rules present; mirror-lock\npolicy present; page-order authority declared).panel_count_band_aligned — panels-per-page in band per target tier {mvp:(2,2), demo:(4,6), longform:(10,12)} (in-range = 5, off-by-one = 3, further = ≤2), AND the TOTALS line reconciles (Σ\npanels-per-page == panel count; NEW + reused == total).continuity_chain_well_formed — the MOTIF STATE TABLE has one row per panel; links have no\ndangling / out-of-order / cycle; every per-panel motifs field agrees with its table row.STRUCTURAL HARD VETO (forces revise regardless of reviewer scores OR Codex — these are NOT scored\n≥4 dims, they are the gate's computed structural vetoes; \"structural failures cannot be voted-around\"):
_unresolved_asset_refs (a referenced asset missing or not locked) — FINAL pass; at the\nprovisional pass declared-but-unlocked refs are expected, and only an UNDECLARED ref (no whitelist entry,\nno complete asset_request) vetoes._policy_violations (a missing/inconsistent global_policies field, incl. no\npage-order-first section)._continuity_breaks (no MOTIF STATE TABLE; a row missing; a motifs field disagreeing)._panel_count_band.MAX, or a post-fall peak).claim_delta value in the exact_parse column or vice-versa).comic.json page order (the orphan-panel class) —\nthe storyboard's page order is the authority; the compiler must conform, not the reverse.Author-time obligations the structural facts above encode (this skill is responsible for them BEFORE the gate\nruns, so the four dims pass and no veto fires): the page-order-first section, the 9-field-per-panel\ncompleteness (DONE panels in the reduced locked form), expected_literals = salient blind-transcribable ASCII\nonly with the deliberate exclusions written out (a sub-gate chip smuggled in is a baked_vs_gate_contract\nbreak), the consolidated one-canonical-owner-per-token asset list (no un-normalized alias), the text-mode/\nblueprint match (a baked panel with no content_blueprint, or an html panel carrying in-frame literals,\nis invalid), and the ✅ APPROVED by user <date> provenance stamp from the outline layer.
Author every panel_spec so the downstream blueprint + spiral never refuse it. Mind the payload-vs-runtime\npath split (do NOT conflate them — see schemas/node_schema.json):
panel_spec.payload.content_blueprint — what THIS skill writes: a deterministic description of which\ncanonical SVG asset is instantiated + the exact literal layout (NOT a raw SVG, and never a\ncondition.content_svg key — that prefix exists only on comic.json).blueprint.payload.content_svg — the actual SVG, authored later by\ncomic-blueprint-author from this panel's content_blueprint.comic.json panel condition.content_svg (+ condition.expected_literals) — the RUNTIME fields the spiral\nengine reads (spiral_engine.js at .condition.content_svg); the compiler maps the blueprint into these.Author to that chain so the downstream never refuses the panel:
\ncontent_blueprint (a resolvable blueprint description) so the blueprint author can\nproduce a non-empty blueprint.payload.content_svg → the engine's condition.content_svg. Do NOT plan a\npanel with no blueprint; the engine rejects a null condition.content_svg at runtime.baked figure-panel MUST declare expected_literals (exact numbers/keys, verbatim) in the panel_spec\n— these become the comic.json condition.expected_literals the engine fail-closes on. A scene panel with no\naudited numbers → text_mode: \"html\" (and its text moves to the HTML overlay).schemas/node_schema.json)Reads — one outline_spec (node_id outline:<slug>, must be status: \"locked\"): its\nnarrative_beats, beat_table, motif_arcs, design_constraint, title/logline/tagline, and the\n*_asset_ids whitelists. Also reads the wiki banlist of prior failed orderings (for the 抽卡 novelty score).
Writes (status under_review → locked after the gate):
storyboard_spec (node_id storyboard:<slug>) — payload required:\nsource_outline_id, page_order, page_ids, panel_ids, global_policies, motif_ledger_id, consolidated_asset_requests. This is the envelope; page_order is the fixed authoritative ordering.motif_ledger (node_id motif:<slug>) — payload required: source_storyboard_id, columns, rows, invariants, mirror_locks. columns = the MOTIF_TABLE_COLUMNS; rows = one object per panel;\ninvariants = the declarative predicates (ddl_monotonic_non_increasing, bounce_single_max,\nmetric_columns_disjoint, motif_vs_env_disambiguation); mirror_locks = the paired constraints.\nThis node is READ by comic-continuity-audit — the table is the\nper-panel ground-truth checklist that turns its cross-model audit from \"infer the intent\" into \"verify\nagainst the row\" (and makes absence ≠ drift design-aware).panel_spec × N (node_id panel:<slug>) — payload required: source_storyboard_id, page_id, panel_id, sequence_index, page_type, world, asset_ids, text_mode, expected_literals, content_blueprint, bubbles, side_narration, motifs. (cover/finale panels additionally carry safe_zones + the endcard\nfields.) Every panel_spec — including DONE/reused — carries ALL the required payload fields above (the\nreduced 3-field form is the STORYBOARD_DRAFT.md convenience only, never the wiki node).continuity_constraint × M (node_id cont:<slug>) — payload required: family, constraint_text, applies_to_panel_ids, source_node_id, active — for a constraint the gate should check\nper-panel as a standalone predicate (e.g. one per mirror lock).Author edges (every type MUST be in validate_wiki.py EDGE_TYPES or the wiki hard-fails — there is no\ncontains / generated_from / references_asset in that set):
derived_from (storyboard_spec → outline_spec) — \"this storyboard was authored from that outline\".uses_motif_ledger (storyboard_spec → motif_ledger) — the ledger this storyboard owns.uses_asset (panel_spec → each whitelisted locked asset it consumes); for a newly-requested asset\n(one filed via asset_requests, not yet built) emit plans_asset (panel_spec → asset) instead.storyboard_spec.payload.panel_ids / motif_ledger_id fields and each panel_spec.payload.source_storyboard_id.All verdict-bearing runtime edges (attempt_of/reviews/decides/rollback_of) belong to Phase 2/3, not\nhere. Trace every gate round to trace.jsonl (review-tracing).
The canonical exhibit is examples/comic_m3_audit/story/STORYBOARD_DRAFT.md\n— the full real artifact: a 24-panel / 19-page pixel-art comic of \"ARIS handed over 24 hours\". Copy its\nexact shape:
storyboard layer draft v1 — revised per codex cross-model critique (verdict REVISE applied); ✅ APPROVED by user 2026-06-10 (\"批了\") — asset-library build started. The user-first gate is honored before any asset work.P00_cover[S01·full-bleed] → P01_b02[S02·single·seam-split] → P02_b03[2-up S03(L warm)|S04(R dark)] → P03_b04[S05] → P04_b05[S06] → P05_b06[vertical-3 S07/S08/S09 big] → P06_b07[2-up mirror S10(L)|S11(R)] → P_B08_0[S12a NEW] → P_B08_1..4[S12–S15 reuse] → P02_b08[grid2x2 recap S12–S15 kept] → P_B09[S16 mirror-of-S11] → P_B09_5[S16b big-frame star-map] → P_B10[S17] → P_B11[2-up seam S18(L)|S19(R)] → P_B12[2-up S20(L)|S21(R seam)] → P03_end[S22 endcard]. TOTALS: 24 panels · 19 pages · 20 NEW bakes +\n4 reused DONE (S12–S15). (19 pages is the AUTHORED plan — the compiler folds the endcard into\nP_B12.closing.image, so the compiled comic.json ships 18 pages; see §Constants PAGE_TYPES mapping.)Panel | Mug | DDL | Tok|yo | exact_parse | claim_delta | bounce. Patterns to copy: the monotonic-DDL fix\n(S06 T-18:40 so it stays > S07's 18:30); split metric columns (S10 exact_parse 0.71→0.66 with\nclaim_delta —; S14 claim_delta +6.2→+1.4 with exact_parse —); bounce uniqueness (S02 是全片唯一 MAX;\nS12a = attempted celebration, ARRESTED before becoming a jump; S21 = smallest bounce); motif-vs-env\n(the </> mug tagged env, never tracked).expected_literals: [\"DDL\", \"T-24:00:00\"], the side_narration trait tag, and the motifs line restating\nthe table row.status + side_narration + motifs (as baked).expected_literals to salient blind-transcribable ASCII only, and write out the deliberate\nexclusions (sub-threshold chips, decorative non-ASCII, skill names).asset_request.MAX bounce, add a post-fall peak, or co-mingle the two metric\ncolumns — each is a hard veto.side_narration only.reviewer-independence — the 抽卡 Codex call and the cross-layer gate get file paths only (the locked outline, style bible, banlist) behind an === EXTERNAL CONTEXT (advisory) === fence, never this agent's interpretation; the brainstorm family ≠ the gate family.acceptance-gate — the loop can DRIVE but can't ACQUIT; user approval of the outline is the hard precondition gate (never proceed without it); the storyboard gate (different model family) acquits the layer.reviewer-routing — Codex at xhigh (NO model pin — follows the local codex config, currently gpt-5.6-sol); Gemini auto-gemini-3; never downgrade the effort tier (effort never lowers reviewer quality).artifact-integrity — the agent that authors the storyboard does not judge its own correctness; the MOTIF STATE TABLE is the ground-truth spec a fresh reviewer verifies, never the author.review-tracing — every 抽卡 round + gate round is logged to trace.jsonl so each verdict (and each dropped lineage) is auditable.