Generate a brand-locked, information-dense single-page PDF that lets a reader understand an entire topic - a company, a product, a project, a status update, any subject - in one page. Distills a raw brief into a composed grid of reusable blocks (hero-stat, KPI strip, two-column prose, comparison table, timeline, callout, icon-grid, fact-rail, mini-chart), renders one HTML via Playwright to a single PNG, mechanically guarantees it fits on ONE page via a fit-guard + tightening ladder, then wraps it as a 1-page PDF. Default 16:9 landscape; A4/Letter portrait via flag. Dark brand canvas default; light print-friendly theme via flag. Use when someone needs a dense one-pager / one-sheet / fact sheet / briefing doc that ships as a PDF.
Standardmäßig ist der Prompt ausgewählt, der zuerst die Quelle prüft. Sie können zu einem direkten Befehl wechseln oder eine lokale Kopie herunterladen.
Quelldateien prüfen
Lesen Sie SKILL.md und alle von SkillsMP angezeigten Begleitdateien, bevor Sie sich für eine Installation entscheiden.
Mit Codex oder Claude installieren Kopieren Sie diesen Prompt, fügen Sie ihn in Codex, Claude oder einen anderen Assistant ein und lassen Sie die Skill-Seite prüfen und installieren.
Ein direkter Befehl überspringt den Prüf-Prompt. Prüfen Sie die Quelle, bevor Sie ihn ausführen.
Generate a brand-locked, information-dense single-page PDF that lets a reader understand an entire topic - a company, a product, a project, a status update, any subject - in one page. Distills a raw brief into a composed grid of reusable blocks (hero-stat, KPI strip, two-column prose, comparison table, timeline, callout, icon-grid, fact-rail, mini-chart), renders one HTML via Playwright to a single PNG, mechanically guarantees it fits on ONE page via a fit-guard + tightening ladder, then wraps it as a 1-page PDF. Default 16:9 landscape; A4/Letter portrait via flag. Dark brand canvas default; light print-friendly theme via flag. Use when someone needs a dense one-pager / one-sheet / fact sheet / briefing doc that ships as a PDF.
{"version":"1.2","created":"2026-06-05T00:00:00.000Z","author":"Ability.ai","locked_design":true,"changelog":["1.2: Promoted to the public trinity-skills library — external dependencies vendored (brand_defaults.md replaces the private design-system doc as a swappable DEFAULT; image_concepts.md replaces the cross-skill object library; render_html_to_png.py copied into scripts/), --autonomous headless mode added so the skill is callable as one line, Drive delivery removed, imagery now invoked via /nano-banana-image-generator by name instead of reaching into its scripts, personal contact details removed from the worked example","1.0 - initial: distill -> content gate -> image -> render -> fit-guard -> visual gate -> 1-page PDF; 16:9 default, dual theme, thin accent + optional inline imagery","1.1 - Tier-3 conformance pass: added Completion Checklist + Error Recovery, effort:high, ultrathink marker; fixed the flex no-shrink layout invariant + 3-signal fit-guard found during validation"]}
One-Pager
Purpose
Generate a brand-locked, information-dense one-pager as a single PNG + a single-page PDF. It is the density inverse of /presentation: where a slide carries one idea at projection scale, a one-pager carries an entire briefing at read-up-close scale. The promise is literal - the reader understands the whole topic from one page, and the output is never more than one page.
The page is composed from a library of reusable blocks (see block_templates.md) laid on a flat row/flex grid. The skill distills a raw brief into telegraphic block content, gates that content for approval, generates a thin brand accent image, renders, mechanically verifies the page fits, and assembles the PDF.
Works for any topic - company overview, product sheet, project brief, investor one-pager, status dashboard, a topic explainer. The block library is generic; the layout preset adapts to the source's intent.
When to use
Right tool when:
Someone needs to understand a whole thing - a company, product, project, topic - from one page
The output should ship as a PDF (one-pagers / one-sheets / fact sheets / leave-behinds are distributed as PDF)
You want it to feel like the same system as its siblings (/presentation, /explain-visually, /microsite) — they share brand_defaults.md
The content is dense and structured (stats, comparisons, timelines, key points) rather than one big idea
Wrong tool when:
The content genuinely needs multiple pages / slides → use /presentation (multi-slide 16:9 deck → PDF)
It's a single social image (one hero + a title/quote) → a single-image social skill
It's a vertical social carousel → a social-carousel skill (4:5, multi-slide)
It's a single structured diagram (one mechanism/flow/comparison) → a diagram skill
The "one page" is really a website / interactive doc → that's a web build, not this skill
What makes this distinct
Approach
Trade-off
Manual one-pager in Canva / Figma / InDesign
Max flexibility, but slow, brand-drift risk, no auto-distillation, no mechanical one-page guarantee
/presentation (multi-slide deck)
Spacious, one-idea-per-slide, many pages - the opposite density profile
/one-pager (this skill)
Single page, dense, brand-locked, distills a brief, mechanically guaranteed to fit on one page. ~$0.07 for the accent image (or $0 with --no-image) + ~1-2 min
State Dependencies
Source
Location
Purpose
Brand defaults
brand_defaults.md (this skill)
Palette, fonts, weight rule, prose-colour rule, imagery lineage - read every time. §0 first: the values are swappable defaults, and the consuming workspace's own design doc wins over them
Block templates
block_templates.md
Canvas, dual theme, dense type tier, block library, layout presets, density budget + fit rules - read every time
Measures whether the page overflows; drives the tightening loop (LOCKED single-page guarantee)
Image generation
the /nano-banana-image-generator skill, invoked by name
Thin accent band + any optional inline image
PDF builder
scripts/build_pdf.py
Single PNG → correctly-sized 1-page PDF
The one-pager reuses the brand design system rather than maintaining its own. The only one-pager-specific specs are in block_templates.md: the single-page canvas, the light theme, the dense type tier, and the block library.
Type-scale exception (LOCKED): the design system's 26px floor is calibrated for mobile-feed glanceability. A one-pager is read up close (laptop full-screen / printed in hand), so it uses a denser tier (60px title, 24px body, 17px source floor) defined in block_templates.md. This mirrors how /presentation took the opposite exception (bigger, for projection distance). Everything else from the design system applies verbatim - fonts, weight rule, accent discipline, no-emoji, prose-colour rule.
Prerequisites & cold start
Script paths in this file are relative to this skill's directory — run them from there,
or prefix with the directory the skill was injected into.
Requirement
Needed for
Missing →
python3 + Playwright/Chromium
rendering (Step 7) and the fit-guard (Step 8)
Report the install command (pip install playwright && playwright install chromium); the single-page guarantee cannot be claimed without the fit-guard, so do not claim it.
PIL/Pillow (pip install pillow)
PDF assembly (Step 11)
Report the install command; the PNG is still a valid deliverable.
GEMINI_API_KEY (fallback GOOGLE_API_KEY)
the accent band + any inline imagery
Not fatal — proceed as if --no-image was passed, and say so in the delivery report. Held by the /nano-banana-image-generator child skill, which fails naming the key.
Generation cost: $0.067 for the accent band, +$0.067 per optional inline image.
--no-image (or no key) = $0.
A brief, a file path (.md/.txt), pasted notes, or a structured block list. Can be messy - this skill distills it (see Content model). If the source is a one-line topic with no substance ("a one-pager about us"), the skill asks for the raw material BEFORE distilling - it does not fabricate facts.
--format
16:9 (1920×1080)
16:9 screen/dashboard (default), a4 portrait print, letter US portrait print, a4-landscape print dashboard. See the formats table in block_templates.md.
--theme
dark
dark = the black canvas of the default palette. light = warm-paper print-friendly variant. Both are swappable per brand_defaults.md §0.
--intent
auto
Picks the layout preset. auto infers from the source; override with company (Preset A), dashboard (Preset B), or briefing (Preset C).
--no-image
(off)
Skip ALL image generation (no accent band, no inline). Pure typographic + data density. $0, faster.
--autonomous
(off)
Headless mode for scheduled/unattended runs: skip both approval gates. [NEEDS:] facts are omitted, never guessed, and listed in the final report. A bare-topic source is a failed run — report it and produce nothing.
--output-dir
one_pager_{slug}/
Where the HTML, PNG, image sources, and PDF land. Relative to the working directory unless given absolutely.
Content model (LOCKED): distill, then gate
The skill accepts a raw brief and distills it into the dense one-pager format. Distillation = selecting blocks, writing telegraphic content within the density budget, and assigning each block to a row. Distillation condenses and structures the user's material; it never invents facts. Numbers, names, claims must come from the source - if a block needs a fact the source doesn't provide, the skill leaves a [NEEDS: …] marker and surfaces it at the content gate rather than guessing.
If the source is already structured as explicit block content (the user did the condensing), the skill detects this and styles it verbatim - it only distills when handed raw material.
Why distill-then-gate: the most expensive failure is generating the accent image + rendering + fit-looping, only to find the wording was wrong or a number was invented. The content gate (Step 4) makes copy iteration free and ensures the visual pipeline runs on locked, fact-checked content.
Headless mode (--autonomous)
This skill is callable as one line by a schedule, an orchestrator, or another agent
(the fleet's playbook-call convention). Invoked that way there is nobody to answer a gate,
so --autonomous must be passed and both gates are skipped. In that mode:
[NEEDS:] gaps are dropped from the page and listed in the final report — never
filled by guessing. Invented facts are the one failure this skill must never produce.
A bare-topic source with no substance is a failed run: report the failure, produce
nothing. It is not an invitation to write the content yourself.
Everything else — distillation, imagery, render, fit-guard, PDF — runs unchanged. The
fit-guard in particular is never skipped; it is the single-page guarantee.
Process (transactional)
ultrathink - this playbook carries non-trivial reasoning: distilling a brief within a fixed density budget (Steps 3-4) and converging the fit-guard tightening ladder (Step 8). Reason carefully at each gate and before each re-render rather than tightening blindly.
Step 1: Read state
Read brand_defaults.md — §0 brand override first, then palette, fonts, weight rule, prose-colour rule. If the workspace has its own design/brand document, port its tokens in and let it override every default.
Read block_templates.md (canvas, dual theme, dense type tier, block library, presets, density budget + fit rules)
Plan the page as an ordered list of rows of blocks. Assign each block a priority (must-keep / trimmable / nice-to-have) - this drives the tightening ladder later. Presets are starting points; swap or drop blocks to fit the actual source.
Step 3: Distill content into the blocks (respect the density budget)
Write the telegraphic content for each block. Hold to the density budget in block_templates.md: ~450 words total, 1 hero-stat, ≤6 KPI cards, ≤2 prose blocks (≤70 words each), ≤6 bullets/list, ≤2 callouts, ≤5 timeline steps, ≤6 body rows (16:9).
If the source has more than fits, KEEP the highest-signal facts (the ones that serve "understand it in one page") and cut the rest - don't plan to shrink the font. Mark any fact the source doesn't supply as [NEEDS: …].
Step 3.5: Density lint
Count words and blocks against the budget. If over, tighten the content now (telegraphic rewrites, drop the weakest block) before showing the user. The page should be plausibly one-page BEFORE the gate.
Step 4: [APPROVAL GATE 1 — CONTENT] Confirm the distilled text + flag any gaps (skipped with --autonomous)
LOCKED first gate. No image generation, no rendering, no PDF until the content is approved. This is where invention-drift and missing facts get caught.
Present the full distilled content as plain markdown, block by block, in reading order, so the user can read the whole page in one scroll. Surface any [NEEDS: …] markers prominently. Example shape:
ONE-PAGER · {slug} · format 16:9 · theme dark · preset: company
HEADER
Eyebrow: Company One-Pager · 2026
Title: Ability|.ai|
Value-prop: Deploy an entire autonomous department in the cloud - not just an agent on your laptop.
Wordmark: trinity.ability.ai
HERO-STAT
72K · YouTube subscribers · +18K in 90 days
KPI STRIP
3.2× ARR growth (YoY) · 94% net retention · 40+ enterprise pilots
TWO-COL PROSE — "The problem / the solution"
... (70 words) ...
FACT-RAIL — "At a glance"
Founded 2024 · HQ Remote · Stage [NEEDS: funding stage] · License Open source
ICON-GRID — "Capabilities"
→ Persistent memory · Multi-agent orchestration · Full governance
TIMELINE — "Roadmap"
Q1 Open source · Q2 Agent Hub · Q3 Enterprise (SOC2 + SSO)
FOOTER + CTA
{Author} · {org} · {year} → {call to action}
Then use AskUserQuestion:
"Approve this content for the {slug} one-pager? Anything to fix - including the [NEEDS] gaps - before I generate the image and render?"
Options:
Approve - render it → proceed to Step 5
Edit a block → revise that block's text (and fill any [NEEDS]), re-show this gate
Cancel → stop, spend nothing
Once approved, write the locked content to {output-dir}/source.md. DO NOT generate images, render HTML, or build a PDF until this gate returns "Approve".
Step 5: Generate imagery (skip if --no-image)
Generate the thin accent band (the brand signature). Match the topic to a concept in image_concepts.md and use its prompt grammar. The band is a wide strip, so request a wide aspect.
Invoke /nano-banana-image-generatorby name, passing the built prompt, the output path {output-dir}/accent_band.png, and a 16:9 aspect. Never call a sibling skill's internal scripts — the child owns its own interface, and its credential handling comes with it.
For any approved inline media block, generate or place that image too (a 1:1 aspect for a contained card, or embed an existing PNG you were given). Verify each image is textless. Prompt sanitization (LOCKED): never use internal double quotes in the prompt string - use single quotes, parentheses, or paraphrase.
Step 6: Author the single HTML
Fork the canvas + theme + block CSS from block_templates.md:
Set the canvas dims for --format and data-theme for --theme on <html>.
Compose the approved blocks into rows per the preset.
Embed images via file://{output-dir}/accent_band.png (and any inline media).
Use the dense type tier for the chosen format. Apply the prose-colour rule and weight rule.
Save to {output-dir}/{slug}.html.
Step 7: Render the PNG
Render at device-scale 2 (dense small type must stay crisp):
Exit 3 (overflow) → apply the tightening ladder from block_templates.md in order, then re-author + re-render + re-check. Repeat until it fits:
Telegraphic rewrites of the longest prose/bullets
Reduce row gap (28→22→18) + card padding (22/26→16/20)
Drop the lowest-priority block
Last resort: step the type tier down one notch (never below the 17px source floor)
The page is not allowed to clip or spill. A render only advances past this step when check_fit.py returns fits: true. Note in the final report what (if anything) was tightened.
Step 9: Show the rendered PNG
Use the Read tool on {output-dir}/{slug}.png so the user sees the full page inline. Report that the PDF has NOT been built yet - awaiting visual approval.
Step 10: [APPROVAL GATE 2 — VISUAL] Confirm it looks right (skipped with --autonomous)
LOCKED second gate. No PDF until the rendered page is approved. Content approval (Gate 1) does not imply visual approval - a typo-free page can still render with an off accent image, an awkward block balance, or a layout that reads poorly.
Use AskUserQuestion:
"Here's the rendered one-pager. Approve it for PDF, or is there a block / the accent image / the layout to fix first?"
Options:
Approve - build the PDF → proceed to Step 11
Regenerate the accent image → re-run Step 5 for the band, re-render, re-show, re-ask
Edit a block → loop back to Step 6 for that block, re-render (+ re-run fit-guard), re-ask
Switch theme / format → re-author for the new theme/format, re-render, re-ask
Cancel → stop, don't build the PDF
DO NOT call build_pdf.py until this gate returns "Approve".
Re-author the row, re-run fit-guard (it may trigger the ladder), re-assemble
Make it denser / sparser
Adjust the density budget for this page, re-distill, re-gate
Step 14: Persist
Keep the full {output-dir} (HTML, PNG, image sources, PDF, source.md) as the reproducible record.
Delivery beyond the local directory (upload, send, publish) is the calling agent's job, not this skill's.
Completion Checklist
Verify before declaring the one-pager done (the transactional close):
Source classified; if it was a bare topic, raw material was gathered - no fabricated facts
Content distilled within the density budget; every [NEEDS: …] gap filled at Gate 1
Gate 1 (content) returned Approve; source.md written to the output dir
Imagery generated and verified textless - or --no-image honored
HTML authored with the correct format dims, data-theme, and dense type tier
PNG rendered at device-scale 2
Fit-guard passed (check_fit.py → fits: true, exit 0) - the page is mechanically confirmed to be ONE page
Gate 2 (visual) returned Approve
PDF assembled at the correct DPI (16:9 → 192, a4/letter → 300)
Output dir retained (HTML, PNG, image sources, source.md, PDF); Drive upload done if requested
Final report delivered: PDF path + page size, PNG path, format/theme/preset, fit-guard notes
Error Recovery
Failure
Recovery
Source is a bare topic with no facts
Halt at Step 1.5 and ask for raw material. NEVER fabricate facts to fill the page.
Nano Banana image generation fails
Retry once. If it still fails, fall back to --no-image (render text-only), note the missing accent band, and offer to add it later. Never block the deliverable on imagery.
Playwright / Chromium render fails
Confirm Playwright + Chromium are installed, then retry the render. If it persists, report the exact error - do NOT assemble a PDF from a stale or partial PNG.
Fit-guard won't converge after the full tightening ladder
The content genuinely exceeds one page. Drop the lowest-priority block(s) and return to Gate 1 to re-confirm the trimmed content. NEVER ship a clipped or overlapping page - the single-page guarantee is non-negotiable.
Pillow missing at the PDF step
pip install pillow, then retry build_pdf.py - the PNG already exists, so no re-render is needed.
Drive upload fails
The local PDF/PNG are the deliverable and are safe. Report the upload error and retry the upload step alone - do not regenerate the one-pager.
Output
Return:
The single-page PDF path - the deliverable - with its physical page size
Walk the tightening ladder (Step 8) - cut content first, type-tier last; never clip
Page reads as a "wall of text"
Too much prose, not enough blocks
Convert prose → KPI cards / bullets / fact-rail / comparison; hold ~450 words
Looks sparse / empty
Too few blocks for the canvas
Add a KPI strip, fact-rail, or timeline; or switch to a portrait format which fills differently
Light theme looks washed out
Used --ink-2 for body prose
Prose is ALWAYS --ink (near-black on light); gray is chrome only
Accent band overpowers the page
Band too tall or image too bright
Keep band ≤120px (16:9) / ≤90px (portrait); the --img-filter + .ab-overlay mute it
Two colors creeping in (green/red deltas)
Off-palette second hue
Use --red for the ONE notable delta, --ink-2 neutral for the rest
PDF prints at the wrong physical size
Wrong --dpi
16:9 → 192, a4/letter → 300
Related skills
/presentation - the multi-slide sibling (16:9 deck → PDF). Use it when the content needs more than one page.
A diagram skill (if installed) - if one block needs a real structured diagram, build it at the matching aspect and embed the PNG as an inline-media block (B11).
A single-image social skill (if installed) - single social images, not dense documents.
