Judge and reshape the story told by an entire paper figure deck. Use when writing or revising a paper to derive a grounded brief from the manuscript and captions, review the full deck as a handling editor, and hand an ordered figure arc to figure-composer.
Tasks
Derive a paper brief from the reviewed manuscript/abstract and captions using the bundled paper_brief schema, Ajv 2020 validation, and a corrective retry loop, stopping on invalid output
Review the full figure deck as a handling editor via narrative_review_task with one delegated reviewer, capturing hook_verdict, arc, figure_moves, missing_panels, kill_list, and boldest_defensible_fig1
Hand each changed figure's composition specification to the figure-composer workflow in the root agent, including claims, moved panels, data Artifact Versions, kill actions, width_mm, and delegate prefixes
Rebuild and publish a new deck in arc order from verified composite Versions and carry the rebuilt deck Version identity into the next review round
Inputs
manuscriptVersionId: immutable manuscript Artifact Version and its reviewed text (abstract-only allowed)
abstractText: reviewed abstract text for bounded brief reasoning
captionsVersionId: immutable captions Artifact Version and reviewed per-figure caption/claim text
deckVersionId: immutable deck Artifact Version containing every current figure in review order
rulesVersionId: immutable design-rules Artifact Version used only as reference
figureDataVersionIds: immutable data Artifact Versions grouped by figure
figureWidthMmByFigure: reviewed positive venue width per figure
Outputs
A validated paper brief (pitch, vision, audience, most-arresting asset, figure claims)
A schema-valid narrative review with hook_verdict, arc, figure_moves, missing_panels, kill_list, and boldest_defensible_fig1
A complete changed-figure composition queue with reviewed claims, moved panels, kill actions, data Version references, width_mm, and delegate prefixes
Verified composite figure Version IDs recorded after producer/reviewer agreement
A rebuilt published deck Artifact Version in arc order, with its immutable rebuiltDeckVersionId retained for the next round
Limitations and checks
Full manuscripts commonly exceed the 64 KiB prompt limit; prefer a reviewed abstract and never silently truncate source text
host.llm does not enforce a caller-provided schema; validation and corrective retry are the caller's responsibility
Delegated children cannot call host.delegate, so the whole figure-composer must not be delegated
Every brief, review, arc, move, omission, and proposed analysis is model-generated and requires human review; generated text must never be described as manuscript evidence or source data
A missing-panel analysis must actually run and publish as an Artifact Version before use; fail closed otherwise
Do not infer affected figures from free text or apply rejected recommendations
Confirm the paper brief passed Ajv 2020 validation against paper_brief_schema() and that stopReason was 'end_turn'; reject on invalid corrective retry rather than guessing missing fields
Verify each delegated reviewer child completed with a schema-valid structuredOutput before using hook_verdict, arc, figure_moves, missing_panels, or kill_list
AI Search
Find projects, verify facts, compare options, or turn a complex need into an actionable plan
Try a searchA click only fills the search box; you stay in control