OPEN-SOURCE AGENT SKILL · V2

One deck system. Your brand.

Turn structured research into a validated presentation website. Choose one of four visual systems or supply your own without rewriting the renderer.

No account. No telemetry. No API key.

Carbon Forge 01 / 04

OPERATING DECISION

Evidence is only useful when it resolves to a decision.

RECOMMENDATION BUILD
4 built-in profiles Custom brand contract Isolated builds Schema validation MIT licensed

Your presentation system should not mistake one house style for a law.

Most deck generators bind content, brand, and renderer together. Change one and the whole system becomes a fork.

V2 separates them. The report stays structured. The brand becomes data. The renderer is copied, adapted, tested, and built in isolation.

THE WORKBENCH

Change the system, not the content.

Select a preset. The preview updates with its functional palette, typography, wordmark, and operating tone.

05
Bring your own brandMap a guide, website, logo, colors, or fonts into a validated profile.
Carbon Forge DECISION BRIEF

WHY IT MATTERS

Make the recommendation visible.

Separate source evidence from judgment so the audience can inspect both.

STATUSREADY RISKVISIBLE

Built for technical reviews, operating briefs, and implementation decisions.

The renderer stays clean. Every deck gets its own build.

The skill follows a fixed sequence so visual changes cannot silently weaken validation or contaminate another presentation.

  1. 1
    Author the report

    Structure sources, findings, judgment, and next actions in report JSON.

  2. 2
    Resolve the brand

    Select a preset or normalize supplied brand material into ten color roles and three font roles.

  3. 3
    Create an isolated renderer

    Copy the bundled renderer into a new marked output directory. The base remains untouched.

  4. 4
    Apply and validate

    Update tokens, metadata, wordmark, schema, report route, and active manifest.

  5. 5
    Test and build

    Run report validation, unit tests, TypeScript checks, and a production build before sharing.

WHAT SHIPS

A skill you can inspect.

The public repository includes the full mechanism, not a prompt that points back to a private machine.

Agent skill
SKILL.md, focused references, scripts, templates, and an uploadable zip
INCLUDED
Renderer
Self-contained Vite and TypeScript presentation runtime
INCLUDED
Brand system
Four presets plus a documented custom profile contract
INCLUDED
Verification
Schema validation, content budgets, unit tests, typecheck, and production build
INCLUDED
Hosting
Static output compatible with Vercel and other static hosts
OPTIONAL
Credentials
No keys, accounts, or private services required for local generation
NONE

Built for

  • Operator briefings that must end in a decision
  • Research reports that need live presentation mode
  • Consultants working across several brands
  • Teams that want validation before visual polish

Not built for

  • PowerPoint or Keynote export
  • Drag-and-drop slide editing
  • Inventing proof, metrics, or testimonials
  • Replacing design judgment with a theme switch

From download to live deck, step by step.

The guide covers installation in Claude, Codex, and Cursor, report authoring, all brand paths, local verification, safe rebuilds, deployment, privacy, and troubleshooting.

Read the complete guide
01cp assets/report-template.json my-report.json
02python3 scripts/prepare-branded-renderer.py
03npm install
04npm run validate && npm test
05npm run typecheck && npm run build

Public by design. Private by default.

The repository contains synthetic sample data, no credentials, no telemetry, and no dependency on Paul’s filesystem. Your report stays on your machine unless you choose to deploy it.

Read the security model

Questions worth answering.

Does it require Claude?

No. The package follows the Agent Skills format and the adapter is a standard Python script. Claude, Codex, Cursor, or a human with a terminal can use it.

Can I use my own brand?

Yes. Supply a JSON profile with ten functional colors, three font roles, a wordmark, optional tagline, and explicit visual rules. The adapter validates it before copying or modifying the renderer.

Why does every deck get a copied renderer?

Branding changes the schema, validator, CSS, fonts, wordmark, and metadata. Isolation prevents one deck from changing another and gives every build a clean rollback point.

Does it host my content?

No. It builds locally. You decide whether to keep the output private or deploy the static site to a host such as Vercel.

Can it export a PowerPoint file?

No. The output is a responsive presentation website. PowerPoint and Keynote export are outside the current scope.

Is TabSquirrel included?

Yes. TabSquirrel is one of the four built-in profiles, alongside Carbon Forge, Executive Signal, and Editorial Studio.

Choose a brand. Keep the evidence. Build the deck.

The public skill is ready to inspect, download, and adapt.