Infographic

Infographic (alias /visual-proof)

Transforme un extrait d'étude en un visuel mémorable — poster, slide, thread hook. Aucune recherche, aucun WebSearch.

Also invoked as /visual-proof — same skill, same outputs.

When to use (vs other skills)

| Skill | Responsibility | Does NOT | |-------|----------------|----------| | /layout-html | Article HTML + 5–10 SVG inline | Raster poster, external prompts | | /infographic | This skill — one memorable visual from existing text | New sources, multi-page layout | | /deep-research | Discovery + synthesis | Visuals | | /bear-hours | Topic framing | Visuals |

Optional pipeline (never auto-chained):

/bear-hours → notes/dense-read → /infographic (optional shareable visual)
/deep-research → /infographic (hook) OU /layout-html (rapport complet)

Hard rules

1. No research — never WebSearch / WebFetch. Work only from provided text/files.
2. One visual — single PNG or single prompt export per run (iterations → visual-proof.v2.png).
3. Faithful to source — never invent sources, numbers, quotes, or claims absent from input.
4. Spec always — write visual-proof.spec.md before generate or prompt export.
5. Privacy strip — apply references/privacy-strip.md before spec finalization.
6. Language — on-image text matches source (lang in spec).
7. ≤7 labels on canvas — image models garble text; if more needed → suggest /layout-html.
8. Style first — ask style via AskUserQuestion before reading content (unless smart-skip).
9. No silent brief edits — mention output paths in chat; do not append to brief.md without consent.
10. Metaphor + hero — every spec names one concrete visual metaphor (a drawable scene, never "boxes and lines") and one hero element that dominates the canvas. See Step 2.

Step 0: Style — AskUserQuestion (always first)

Use AskUserQuestion — one call, wait for answer before continuing.

Title: Infographic — Style

Hard rules:

• Never list style options as plain chat bullets — always AskUserQuestion
• STOP after the call; wait for answer
• Smart-skip — if user already named a preset or described custom style (≥5 words), skip (see references/palettes.md)

Question: "Quel style visuel pour cette infographie ?"

Options (exactly 4 — AskUserQuestion max; custom style goes through the automatic "Other" option):

| id | label | |----|-------| | warm | Warm Bear — earthy, amber & vert, School of the Bear (défaut) | | academic | Academic Blue — rapport recherche, bleu pro | | minimal | Minimal Mono — technique, neutre, épuré | | creative | Creative Spicy — coloré, original, un peu WTF |

If user answers via "Other": treat the free text as a custom style. Map to palette per references/palettes.md. One short follow-up only if too vague.

If preset selected: load palette from layout-html/references/presets/{{PRESET_ID}}.md via references/palettes.md.

Default when skipped: warm.

Step 1: Intake — detect input mode

Resolve input in this priority order:

| Mode | Trigger | Source (first with substance) | Default output dir | |------|---------|-------------------------------|-------------------| | A — Research slug | slug or research// | report.md (exec summary if long) | research// | | B — Study slug | studies// | notes.md → research.md → brief.md | studies// | | C — File path | path to .md, .txt | that file | same directory | | D — Pasted text | user pastes in chat | message body | output/layout// | | E — Notes | notes.md present | studies//notes.md | studies// |

Slug derivation (mode D): lowercase-hyphen from title or first heading; ask once if ambiguous.

Guards:

• No resolvable text → ask for slug, path, or paste
• Multiple substantive files in study → AskUserQuestion "Which file is the source of truth?"
• Source 7 precise labels or exact numeric data on canvas → warn user; suggest /layout-html for SVG reliability. Proceed with simplified visual or PROMPT only if user confirms.

Step 4: Build spec — always write

Write {output_dir}/visual-proof.spec.md per references/spec-schema.md.

This file is the audit trail and source for both IMAGE and PROMPT modes.

Step 5: Generate or export

Load references/tool-detection.md. Determine MODE before acting.

Step 5a: IMAGE mode

When: image_gen or GenerateImage available in this session (unless user asked prompt-only).

Before generate: Apply the labeled-infographic verification loop from references/tool-detection.md.

1. Compose prompt from spec + references/prompt-template.md § IMAGE mode
2. If using a prompt-style image tool: pass the composed prompt and aspect ratio (16:9 default; 1:1 if square/Instagram; 9:16 if story)
3. If using a description + filename image tool: write visual-proof.png in the output dir
4. Verify (fidelity loop, see references/tool-detection.md): read generated image back — check text legible and label count/palette/hero match the spec; one targeted retry, then fallback PROMPT
5. Write {output_dir}/visual-proof.png
6. Write {output_dir}/visual-proof.alt.md — 2–3 sentence alt text + caption for posts

Iterate: image_edit on existing PNG → visual-proof.v2.png

open "{output_dir}/visual-proof.png"   # macOS, IMAGE mode

Step 5b: PROMPT mode

When: No image tool, user requested prompt-only, or IMAGE failed.

1. Fill references/prompt-template.md → write {output_dir}/infographic-prompt.md
2. Include blocks: Universal, Midjourney, DALL·E, Flux/SD
3. Print Universal block in chat for immediate copy-paste

Step 6: Deliver

Tell the user:

1. Mode used (IMAGE | PROMPT)
2. Visual type + preset
3. Output paths (spec always; png or prompt)
4. Source file + wedge (if any)
5. Reminder: spec = source of truth; image = illustrative proof, not new research
6. Suggest pairing: /layout-html for full article — do not create proof.md unless user asks

Do not silently edit brief.md. Optional on consent: add ## Visual proof section with paths and date.

Escape hatches

| User says | Action | |-----------|--------| | slug only | Try research// then studies// | | file path | Mode C — outputs alongside source | | pasted text | Mode D — output/layout// | | overwrite | Replace PNG/prompt without confirmation | | warm / academic / etc. in message | Smart-skip style question | | prompt only / pas d'image | Force PROMPT mode | | square / Instagram | aspect 1:1 | | story / vertical | aspect 9:16 | | wants multi-diagram article | Suggest /layout-html | | attaches generated image + iterate | image_edit path | | wants research but no notes | Suggest /deep-research or /bear-hours first |

Self-check before delivery

• [ ] No WebSearch / WebFetch
• [ ] All labels/numbers traceable to input
• [ ] visual-proof.spec.md written with type_rationale
• [ ] Spec has a concrete metaphor (drawable scene) + hero element
• [ ] Privacy strip applied
• [ ] ≤7 on-image labels, 1–2 words each (or user warned + confirmed)
• [ ] IMAGE: PNG + alt.md OR PROMPT: infographic-prompt.md with 4 blocks
• [ ] Universal block printed in chat (PROMPT mode)
• [ ] Clear differentiation from /layout-html in delivery message
• [ ] IMAGE mode: browser opened (open on macOS)

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: loursCreatif
• Source: loursCreatif/ours-stack
• License: MIT

Install and usage instructions live in the source repository linked above.