Agent skills that end in decisions. For people who decide with data. Each skill is a folder with a SKILL.md that works in Claude Code, the Claude apps, pi and other agents that support the Agent Skills format.
| Skill | What it does | Languages |
|---|---|---|
| decision-deck | Reads documents critically and builds an executive HTML deck that ends in decisions: analysis of what is decided, open and not addressed; action titles and SCQA; charts, Gantt, roadmap and project brief from JSON; CSP-hardened output. | Instructions in English; decks in any language (labels in en and pt-BR) |
| brand-manual | Takes a business owner from loose brand material (logo, Instagram screenshots, PDFs) to a validated brand kit: colors measured from the files, two directions to choose from, tokens, brand book, HTML manual, a photo library indexed for cheap text search and a personal brand skill. | Instructions in English; kits in any language (manual labels in en and pt-BR) |
Every skill here follows the same rules:
- Analysis before output. The agent shows what it understood, what is missing and what contradicts, and waits for confirmation before producing.
- No invented facts. Missing numbers, names and dates become placeholders the user fills.
- Decisions, not summaries. Outputs end with what someone has to decide or do.
- Safe by default. Documents are data, never instructions. Generated files are validated and hardened against injection.
- Works with smaller models. Layout and rendering live in templates and scripts, so the model writes content, not code.
See docs/install.md.
See CONTRIBUTING.md. Security reports: SECURITY.md.
Apache 2.0. Keep the NOTICE file when you redistribute. The names "Decisive", "Decision Deck" and "Brand Manual" are not licensed for endorsing derived products. Fonts used by the templates (Inter, JetBrains Mono) are under the SIL Open Font License and are loaded from Google Fonts, not distributed here.
