메뉴
Data Journalist Agent (Data2Story) — orchestrator: turn a dataset into a blog. Runs detective → analyst → editor → designer → programmer → auditor → inspector in sequence. Creates a versioned project folder for each run.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Exhaustively profile a dataset and list ALL possible analyses — distributions, correlations, rankings, trends, group comparisons, anomalies. Reads detective.json for context. Runs after the Detective/Scout, before the Editor. Outputs analyst.json with ana_xx IDs and chart-ready data_tables.
Audit a generated Data2Story blog for build correctness across ALL modalities by ACTUALLY RENDERING it in a real headless browser (when available) — catching blank/0-width charts, broken/oversized media, desktop+mobile overflow, and dead interactions that source inspection cannot see — then having vision agents review the screenshots, playtesting every interactive, and walking the flagship capability contract. Plus VIEW every image for a wrong/AI-faked subject. Fixes layout in place; sends content-correctness problems back to the owning role. Use at Stage 6 after the Programmer assembles index.html and before the Critic; re-runs each revision round. Triggers: a built index.html exists, or you need to confirm charts/media/interactions actually render.
Design the MANDATORY cinematic scroll experience (every blog ships one): a full-bleed, scroll-driven background — primarily from the Scout's verified real imagery — plus the global motion choreography, so the narrative unfolds like a film as the reader scrolls. Owns the BACKGROUND layer + page-level motion (not the Designer's per-section visuals, not the Interaction centerpiece — it stages them inside the scroll). Outputs cinematographer.json (cin_xx scenes). There is no off path: a topic with no sourceable real imagery falls back to a generative-atmosphere or data-driven-spine background, never to a bare column.
Name the piece — re-write the masthead (headline + standfirst + kicker), every section title, and every figure/photo/table caption to a research-driven titling standard, killing the AI-tell patterns (the 'flat statement. flat counter-statement.' two-beat above all) a competent default falls into. Reads editor.md/json + analyst.json + the resolved topic_profile; writes copywriter.json — STRINGS ONLY (masthead{headline,standfirst,kicker}, items{edt_xx:{title}, des_xx:{caption}}), each backed by a real ana_*. Names, never edits: it touches no finding, no number, no data-* id, no layout — so the Verify layer is untouched and the Programmer renders the masthead + figcaptions from copywriter.json verbatim. Runs at Stage 3.5, after the Editor, before the Designer.
Review a finished Data2Story blog against the 5 quality rubric dimensions (visual_design, narrative_pacing, data_method_transparency, claim_data_alignment, insight_value), score each 1-7 with on-page evidence, and emit critic.json with pass/fail + targeted, surgical send-back instructions. Verifies every load-bearing claim/asset against its traceability chain before scoring; applies the caveat-survival, honest-accuracy, and third-party-attribution caps. Does NOT rewrite content — scores and sends back. Use at Stage 6.5 after the Auditor and after verify.py has produced verifier.json; re-runs each revision round. Triggers: a built index.html plus verifier.json exist, or you need to judge whether the article is actually good.
Generate text embeddings via OpenRouter using Qwen3-Embedding-8B.
| name | data2story |
| description | Data Journalist Agent (Data2Story) — orchestrator: turn a dataset into a blog. Runs detective → analyst → editor → designer → programmer → auditor → inspector in sequence. Creates a versioned project folder for each run. |
| argument-hint | ["data path"] |
| allowed-tools | Bash(*), Read, Write, Glob, Grep, Skill, Agent, WebSearch, WebFetch |
Turn $ARGUMENTS into a blog. Orchestrates the roles below in sequence.
Resolve paths before doing anything:
SKILL_DIR = the directory containing this SKILL.md (.../skills/data2story)ARCHIVE_DIR = the ancestor directory that contains skills/ (two levels up from SKILL_DIR, i.e. SKILL_DIR/../..)DATA2STORY_ROOT = parent of ARCHIVE_DIRARCHIVE_DIR; replace them with resolved, quoted paths before running Bash.DATA_NAME = the dataset folder name (e.g. pick_a_card)DATA_DIR = if $ARGUMENTS is an existing path, use that path; otherwise use DATA2STORY_ROOT/data/{DATA_NAME}TIMESTAMP = current time formatted as MMDD_HHMM (e.g. 0401_1618): date +%m%d_%H%M (run in bash)PROJECT_DIR = DATA2STORY_ROOT/project/{DATA_NAME}/blog_{MODEL}_{TIMESTAMP}PROJECT_DIR/, PROJECT_DIR/assets/, PROJECT_DIR/code/Immediately after creating PROJECT_DIR, snapshot the current skills:
mkdir -p PROJECT_DIR/archival
cp -r ARCHIVE_DIR/skills PROJECT_DIR/archival/skills
This preserves the exact skill versions used for this run.
All media tools route through OpenRouter. Set OPENROUTER_API_KEY before any generation call.
Media generation is the Designer's job, so the media tools (text2image, text2video, image2video, text2music, embeddings) live under SKILL_DIR/designer/scripts/openrouter-*/. The full list — default models and exact python3 ... invocations — is in designer/references/tools.json; full per-tool docs are each tool's own SKILL.md under SKILL_DIR/designer/scripts/openrouter-*/.
The pipeline is a single linear sequence that produces a traceable HTML blog from raw data:
DATA → Detective → Analyst → Editor → Designer → Programmer → Auditor → Inspector → final index.html + viewer.html
Run each stage in order. Each stage reads the previous artifact(s) before starting. Do not proceed to the next stage until the current artifact is complete.
Input: DATA_DIR
Output: PROJECT_DIR/detective.json
What: Researches external context — background knowledge, domain history, related findings, why this data matters. Each finding gets a det_xx ID.
Input: DATA_DIR, PROJECT_DIR/detective.json
Output: PROJECT_DIR/code/*.py, PROJECT_DIR/analyst.json
What: Exhaustive quantitative analysis of the data, informed by detective's context. All code saved to code/ as runnable scripts. Each finding gets an ana_xx ID with calculation (file + lines + output) and data_table (chart-ready data).
Input: PROJECT_DIR/detective.json, PROJECT_DIR/analyst.json
Output: PROJECT_DIR/editor.md, PROJECT_DIR/editor.json
What: Editorial decisions — which findings matter, what the narrative arc is, what the blog argues. Each section gets an edt_xx ID with explicit references to ana_xx findings and det_xx context. No visual design.
Input: PROJECT_DIR/editor.md, PROJECT_DIR/editor.json, PROJECT_DIR/analyst.json
Output: PROJECT_DIR/designer.json, PROJECT_DIR/assets/*
What: Data-driven creative visual decisions — how to present each point using charts, images, video, audio, maps, interactives, stat callouts, instances, or text-only treatment when appropriate. The media mix should emerge from the dataset's properties, not from a fixed checklist. The page should be multimedia-rich by default: borrow the visual language from the shared frontend-design skill and use all five channels (chart, image, video, audio, interactive/map) unless a channel's documented fallback would be fabricated or purely decorative. Each visual gets a des_xx ID with data_source pointing to ana_xx data_tables when data-driven. Generates selected assets. No HTML.
Input: PROJECT_DIR/editor.md, PROJECT_DIR/editor.json, PROJECT_DIR/analyst.json, PROJECT_DIR/designer.json
Output: PROJECT_DIR/index.html
What: Implements the final blog in HTML. Applies the theme/accent recorded in designer.json page_rhythm and borrows component + token recipes from the frontend-design skill. Resolves chart data from analyst.json data_tables (NO raw data access). Tags every element with data-edt, data-ana, data-det, data-des attributes for traceability.
Input: PROJECT_DIR/index.html
Output: PROJECT_DIR/index.html (modified), PROJECT_DIR/auditor.json
What: Detects and fixes layout issues (overlap, spacing, alignment) without changing content or design intent. Runs automatically after Programmer to ensure visual elements are properly wrapped and spaced.
Call: Skill auditor PROJECT_DIR
Input: PROJECT_DIR/index.html, all JSON files
Output: PROJECT_DIR/inspector.json, PROJECT_DIR/viewer.html
What: Runs sentence-level traceability verification and generates an interactive viewer. Two steps:
python3 SKILL_DIR/inspector/scripts/verify.py PROJECT_DIR --log-errors
python3 SKILL_DIR/inspector/scripts/generate_viewer.py PROJECT_DIR
Step 1 produces inspector.json (sentence→evidence mapping). Step 2 produces viewer.html (self-contained, works on file:// — no server needed). See inspector/SKILL.md for details.
det_01 ──┐
det_02 ──┤
├──▶ ana_01 (based_on: [det_02]) ──┐
│ ana_02 (based_on: []) ├──▶ edt_01 (findings: [ana_01, ana_02], context: [det_01]) ──▶ des_01 (section: edt_01, data_source: ana_01)
│ ana_03 (based_on: [det_01]) │ edt_02 (findings: [ana_03], context: [det_02]) ──▶ des_02 (section: edt_02, data_source: ana_03)
└────────────────────────────────────┘
Every value in the final HTML can be traced: HTML data-des="des_01" → designer.json des_01.data_source="ana_01" → analyst.json ana_01.calculation.code → verifiable.
designer.json should exercise all five channels (chart, image, video, audio, interactive_or_map). For any channel marked used:false, confirm its documented fallback was genuinely tried and a data-grounded reason recorded in meta.media_decisions. If a channel was skipped for convenience rather than because the data can't support it, send it back to the Designer before the Programmer runs.PROJECT_DIR/assets/ only.PROJECT_DIR/index.html, PROJECT_DIR/detective.json, PROJECT_DIR/analyst.json, PROJECT_DIR/code/*.py, PROJECT_DIR/editor.md, PROJECT_DIR/editor.json, PROJECT_DIR/designer.json, PROJECT_DIR/inspector.json, PROJECT_DIR/viewer.html.