
Infographics
- 108 installs
- 3.6k repo stars
- Updated August 5, 2026
- basicmachines-co/basic-memory
infographics is a Claude skill that generates evidence-grounded PR, changelog, release, and weekly repository images from the Basic Memory PR body via an OpenAI image script.
About
This skill generates repository visuals (PR, changelog, release, and 2-week weekly images) grounded in evidence from the PR body or a changelog evidence pack. It runs an image-generation script that produces image-first editorial scenes and writes them to canonical webp paths. A developer uses it to attach visuals to PRs and releases without hand-writing claims not present in the source.
- Generates PR, changelog, release, and weekly repository images from evidence in the PR body
- Writes to canonical output paths under docs/assets/infographics/ (pr-<number>.webp, changelog.webp, weekly windows)
- Uses an OpenAI image-generation script with a deterministic BM style pool and provenance block
Infographics by the numbers
- 108 all-time installs (skills.sh)
- Ranked #781 of 1,335 Generative Media skills by installs in the Skillselion catalog
- Data as of Aug 5, 2026 (Skillselion catalog sync)
infographics capabilities & compatibility
- Capabilities
- image generation · changelog visuals · release visuals
- Works with
- github · openai
- Use cases
- image generation · marketing
What infographics says it does
Generate repository visuals with evidence-grounded content and canonical output paths.
`--dry-run` is an alias for `--print-prompt`; both print the final prompt and exit without calling OpenAI.
npx skills add https://github.com/basicmachines-co/basic-memory --skill infographicsAdd your badge
Show developers this skill is listed on Skillselion. Paste this into your README.
| Installs | 108 |
|---|---|
| repo stars | ★ 3.6k |
| Last updated | August 5, 2026 |
| Repository | basicmachines-co/basic-memory ↗ |
What it does
Generate evidence-grounded PR, changelog, release, or weekly repository images.
Who is it for?
Producing PR, changelog, release, and weekly images tied strictly to evidence in the PR body or changelog.
Skip if: Text-heavy infographics or hand-written claims not present in the source PR body.
When should I use this skill?
Generating Basic Memory PR, changelog, release, or weekly images from Codex.
What you get
A webp image at a canonical path plus a provenance block recording model, size, theme source, and visual direction.
- pr-<number>.webp
- changelog.webp
- weekly window image
By the numbers
- 10+ BM style categories in the pool
- always a 2-week retro window for weekly images
Files
Basic Memory Images
Generate repository visuals with evidence-grounded content and canonical output paths. The file and marker names still say "infographic" for compatibility, but PR generation is image-first: scene, poster, painting, photograph, cover, tableau, staged artifact, or another editorial visual moment that describes the intent of the PR. PR images are non-gating BM Bossbot artifacts; changelog and release-summary images are manual evidence-pack workflows.
Output Contract
- Base output directory:
docs/assets/infographics/ - PR image:
docs/assets/infographics/pr-<number>.webp - Changelog image:
docs/assets/infographics/changelog.webp - Weekly image:
- This is always a 2-Week Retro window: previous ISO week through current ISO
week (start-week = current-week - 1, end-week = current-week).
- Same year window:
docs/assets/infographics/<year>-w<start-week>-w<end-week>.webp - Cross-year window:
docs/assets/infographics/<start-year>-w<start-week>-<end-year>-w<end-week>.webp
PR Mode
PR mode uses the BM Bossbot summary block as source material. Do not hand-write claims that are not present in the PR body.
1. Fetch the PR body:
gh pr view <number> --json body --jq '.body // ""' > /tmp/bm-pr-body.md2. Generate the canonical asset:
uv run --script scripts/generate_pr_infographic.py \
--pr-number <number> \
--pr-body-file /tmp/bm-pr-body.md \
--theme "<optional visual theme>" \
--provenance-output /tmp/bm-infographic-provenance.md \
--output docs/assets/infographics/pr-<number>.webpIf the PR body contains a managed image theme block, the script reads it automatically:
<!-- BM_INFOGRAPHIC_THEME:start -->
<theme>
<!-- BM_INFOGRAPHIC_THEME:end -->Before spending an image call, test the prompt path locally:
uv run --script scripts/generate_pr_infographic.py \
--pr-number <number> \
--pr-body-file /tmp/bm-pr-body.md \
--theme "<optional visual theme>" \
--output docs/assets/infographics/pr-<number>.webp \
--print-prompt--dry-run is an alias for --print-prompt; both print the final prompt and exit without calling OpenAI.
When no theme is supplied, the script selects a deterministic BM visual direction from the style pool below based on the PR number and Bossbot summary. This keeps repeated PR images from collapsing into the same generic visual.
When the image is generated, also write provenance with --provenance-output <path>. BM Bossbot publishes that managed block into the PR body with these markers:
<!-- BM_INFOGRAPHIC_PROVENANCE:start -->
...
<!-- BM_INFOGRAPHIC_PROVENANCE:end -->The provenance block records the generated asset path, image model, size, quality, image mode, theme source, and selected visual direction. It intentionally does not dump the full generated prompt into the PR body. Treat this block as debugging and creative provenance only; it is not a merge gate.
The PR image is visual support only. The authoritative merge gate is the GitHub commit status named BM Bossbot Approval.
Changelog Mode
Build an evidence pack before writing a prompt:
- diff truth source: merged PR diffs, merge commits, or local reconstructed diffs
- changed-file orientation:
git diff --statplus key file reads - impact ledger: before/after outcomes tied to actual changes
- discard list: misleading titles, reverted work, rename-only churn, speculative TODOs
- chosen image form: poster, scene, tableau, cover, painting, photograph,
staged artifact, or another editorial visual moment
- chosen BM style category: exactly one category from the selection pool below
Read these references before drafting the prompt:
references/prompt-blueprint.mdreferences/style-balance.md
Read the current CHANGELOG.md entries and include the latest meaningful changes.
Style And Category Selection
Select exactly one BM style category per image based on semantic fit. The visual language should be recognizable and tasteful, while staying business-readable.
Create an image-first visual form that communicates the change: poster, scene, tableau, cover image, painting, photograph, staged artifact, or another editorial visual moment. Maps, diagrams, dossiers, charts, and labels can appear as props inside the scene, but do not make a text-heavy infographic.
BM category pool:
- computer science college textbooks: SICP-style diagrams, algorithms lectures,
compiler pipelines, automata, database systems, type theory, operating systems
- classic literature subjects: sea voyages, gothic manors, Dickensian city maps,
Austen social graphs, library marginalia, travel journals
- fantasy/D&D-inspired: quest maps, dungeon keys, guild ledgers, spellbooks,
bestiaries, tavern notice boards; no copyrighted settings
- Music: Metal, Hard Rock, Punk, techno, soul, reggae bands; no pop music, no
direct band logos, album covers, or musician likenesses
- sci-fi: Star Wars inspired knockoff, Spaceballs-adjacent space opera, fleet
routes, mission consoles, contraband manifests; avoid copyrighted characters, logos, or named fictional universes
- Conan the barbarian-inspired sword-and-sorcery: ruined temples, desert routes,
battle standards, ancient maps; no named character likenesses
- Comic books: issue covers, splash pages, action-panel maps, caption boxes,
halftone energy, clean sound-effect typography
- French new wave movies: poster style, stark typography, city route maps,
jump-cut sequencing, high-contrast editorial photography cues
- WWII propaganda posters: home-front public-information poster language,
logistics arrows, ration charts, mobilization maps, bold simplified figures; no real-world party symbols, hate imagery, dehumanizing slogans, or false historical claims
- Italian movie posters: hand-painted drama, bold credits, expressive color,
route-map collage, 1960s or 1970s cinema energy; no direct film titles or actor likenesses
- Shakespeare: stage maps, acts and scenes, dramatis personae, royal courts,
backstage cue sheets
- Greek mythology: temple diagrams, constellation routes, hero's journey maps,
oracle tablets, labyrinths, ship routes
- noir detective boards: case files, red-string maps, typed evidence labels,
precinct wall charts
- NASA mission-control dashboards: launch timelines, telemetry maps, orbital
routes, status boards
- space exploration and astronomy: celestial atlases, observatory charts,
star-field maps, orbital mechanics diagrams, planetary survey routes, telescope annotations, mission trajectories, deep-space timelines
- paintings: abstract painting, classical landscape, Remington-inspired western
action painting, Rembrandt-inspired chiaroscuro, historical mural, stormy seascape, allegorical editorial painting
- classic black-and-white photography: documentary field report, newsroom
archive print, editorial photo essay, street photography, high-contrast darkroom print, contact sheet, civic infrastructure photograph
- 80's action movies: practical explosions, smoky backlit warehouses, neon city
streets, helicopter searchlights, mission dossiers, heroic silhouettes, high-stakes countdowns, painted ensemble posters; no direct actor likenesses, real film titles, franchise marks, or catchphrases
- alchemy manuscripts: transformation diagrams, annotated symbols, recipe-like
process maps, illuminated margins
- brutalist civic planning: transit maps, concrete signage, zoning blocks,
infrastructure diagrams
Selection rules:
- Pick one category only; do not create mixed mashups.
- Pick the most appropriate image form. Prefer an actual scene, poster,
painting, photograph, tableau, or cover over a text-heavy infographic.
- Match metaphor to content, but do not overthink it. The category is a creative
catalyst, not a semantic constraint.
- Use a polished editorial rendering direction: smooth anti-aliased
text, high contrast, clean edges, readable labels.
- Make the category drive the composition through a readable staged moment,
editorial composition, symbolic environment, route, artifact, or visual metaphor.
- Keep the structure literal enough to aid understanding, but not so heavy that
it obscures engineering meaning.
- Give the image generator creative latitude on layout, structure, color palette,
and visual metaphors. Be precise about what content to show, loose about how to show it.
- Do not use copyrighted characters, logos, or named fictional universes. Use
genre cues, knockoffs, and original compositions instead.
Content-First Aesthetic Contract
The meaning must be readable and clearly hierarchical. Everything else is creative territory: image form, layout, visual metaphors, decorative elements, color choices, and category-specific visual language.
Hierarchy:
1. Meaning: what shipped, what changed, and why it matters must be clear. 2. If the image uses text, labels, sections, or evidence bullets, they must be legible. 3. The selected category's visual DNA should drive the composition as a poster, scene, painting, photograph, tableau, cover, or symbolic object arrangement. 4. Do not play it safe. A visually striking image that someone wants to look at beats a correct but boring one.
Hard rules:
- Content sections and labels must be readable when present. Text cannot be
obscured by decorations.
- Do not use lore-heavy copy that competes with engineering or business meaning.
- Every prompt must include a clear image-first composition cue: a staged scene,
poster composition, painting, photograph, symbolic tableau, hero object, mission room, dossier, artifact, route, or visual metaphor.
- Do not over-prescribe exact coordinates or panel geometry; give a composition
backbone and let the model compose around it.
Generation
1. Write the final prompt to a temporary markdown file. 2. Generate with the shared image helper:
uv run --script scripts/generate_infographic.py \
--prompt-file /tmp/bm-infographic-prompt.md \
--output docs/assets/infographics/<name>.webp3. Verify the image exists and is readable before reporting success.
Quality Bar
- Tell a concrete before/after value story, not vague improvement claims.
- Stay understandable for both engineers and non-technical stakeholders.
- Use plain-language section titles and labels when text is present.
- Include clear visual hierarchy: title, staged focal point, symbolic scene,
evidence props, or hero object.
- Avoid invented facts; only use provided source material.
- Favor shipped outcomes over intermediate or reverted work.
- Preserve readability with high contrast, non-tiny labels, and uncluttered
layout.
interface:
display_name: "Infographics"
short_description: "Generate Basic Memory repo infographics"
icon_small: "./assets/icon.svg"
icon_large: "./assets/icon.svg"
brand_color: "#2563EB"
default_prompt: "Use $infographics to generate a Basic Memory PR or changelog infographic with canonical output paths."
<svg xmlns="http://www.w3.org/2000/svg" width="128" height="128" viewBox="0 0 24 24" fill="none" stroke="#111827" stroke-width="1.8" stroke-linecap="round" stroke-linejoin="round">
<path d="M3 12h4l2-6 4 12 2-6h6"/>
<path d="M4 20h16"/>
</svg>
Prompt Blueprint
Convert an evidence pack into a final visual prompt. Be precise about the content and loose about visual execution.
Required Inputs
- Diff truth source summary
- Changed-file orientation summary
- Impact ledger with before/after outcomes
- Discard list for excluded noise
- Chosen image form
- Chosen BM style category
Prompt Shape
Create a polished Basic Memory editorial image inspired by
<BM_STYLE_CATEGORY>. Use a poster, scene, tableau, painting, photograph, cover
image, staged artifact, or another image-first form that best communicates the
intent. Use HD editorial rendering with smooth anti-aliased text when text is
present. Go bold and let the selected category drive the visual language through
original, non-infringing cues.
TITLE:
- "<clear title>"
- "<scope subtitle>"
COMPOSITION:
- Recreate a clear staged moment or symbolic image that describes the PR
intent.
- Maps, diagrams, dossiers, route lines, labels, and artifacts can appear as
props inside the scene, but the output should read as an image rather than a
dense infographic.
- Take creative liberty with layout and styling.
- The hard rule: the meaning must be readable and clearly hierarchical.
- Keep labels plain-language and technical when labels are used.
CONTENT:
1. "<section>"
- <evidence-grounded outcome>
- <evidence-grounded outcome>
2. "<section>"
- <evidence-grounded outcome>
- <evidence-grounded outcome>
METRICS:
- <metric>
- <metric>
STYLE DIRECTION:
- Upscaled editorial, high contrast, anti-aliased text, smooth edges.
- Let the category's visual DNA drive the composition.
- Use genre/category cues only; do not use copyrighted characters, logos, named
fictional universes, direct band logos, album art, or celebrity likenesses.
DO NOT:
- Make text unreadable or let decoration obscure content.
- Render a text-heavy infographic, dashboard, flowchart, timeline strip,
checklist, bullet-list panel, or dense explanatory diagram.
- Use crunchy low-resolution pixel art.
- Invent facts not present in the evidence pack.Writing Rules
- Keep each bullet specific and evidence-grounded.
- Prefer outcome language over implementation trivia.
- Default to three or four sections; never exceed five.
- Give proportionally more space to dominant changes.
- Keep the final prompt short, energetic, and readable.
Style Balance Rubric
Core Principle
Be bold, not confusing. The selected BM style category should structure the visual through a readable image-first composition, not decorate a generic grid. Use an editorial scene, poster, painting, photograph, cover, staged artifact, or tableau that turns the PR intent into a visual moment.
Required Traits
- Anti-aliased typography
- Smooth edges
- High contrast between text and background
- Plain-language section labels
- Clear composition backbone: staged scene, editorial poster, painting,
photograph, symbolic tableau, hero artifact, dossier, mission room, or route embodied as part of the scene
- A single coherent BM style category, expressed through original visual cues
Reject Or Rewrite If
- Content text is unreadable.
- The prompt lacks a composition backbone.
- The prompt over-prescribes exact panel positions or a rigid grid.
- The style leans into crunchy low-resolution pixelation.
- Copy uses lore-heavy references instead of engineering meaning.
- The prompt uses copyrighted characters, logos, named fictional universes,
direct band logos, album art, or celebrity likenesses.
Creative Integration Patterns
- Use category-native map details to organize content: textbook diagrams,
literary journeys, quest maps, tour posters, mission-control routes, stage blocking, mythic constellations, star charts, mission trajectories, case boards, or civic plans as props inside the image.
- Recreate a scene, editorial poster, painting, photograph, cover, artifact, or
tableau instead of sectioned bullets.
- Map engineering metrics to visual counters, route progress, or status boards
only when they naturally belong in the scene.
- Let headers and accents borrow from the selected style.
- Keep atmospheric details behind or around content, never over it.
Related skills
FAQ
Where are images written?
To canonical paths under docs/assets/infographics/, e.g. pr-<number>.webp and changelog.webp.
Can I preview the prompt before spending an image call?
Yes. Run the script with --print-prompt (aliased as --dry-run) to print the final prompt and exit without calling OpenAI.