TIKAZ AI Skills

✨ One collection. Seven workflows. One quality contract.

TIKAZ AI Skills is a monorepo maintained by TIKAZ for Codex and compatible Skill hosts, with contributions welcome. Seven suites live together because useful work crosses boundaries: Context Economy can prepare the smallest reliable handoff, video evidence can become research, research can become a deck, and an approved interface can move into engineering delivery.

Each suite remains independently installable. The collection supplies the shared rules that keep handoffs coherent:

USER GOAL
   ↓
ONE PRIMARY SUITE  ──→  specialist Skills only when needed
   ↓
STRUCTURED HANDOFF ──→  evidence · uncertainty · source · license
   ↓
QUALITY GATE       ──→  render · test · review · artifact proof
   ↓
VERIFIED DELIVERY

This is not an OpenAI-official repository. It is designed, integrated, refactored, and continuously maintained by TIKAZ for real Codex workflows.

🚀 Start with the flagships

Use frontend-design to redesign this dashboard. Write the Design Read,
set the three project dials, and approve a desktop/mobile proof first.

Use video-platform-reader to compare these videos. Keep timestamps,
label evidence levels, and list every claim that remains unverified.

🧩 The seven-suite system

There are 30 installable Skills, including the seven suite orchestrators. Use a suite orchestrator for an end-to-end outcome; install a child Skill when you need only one focused capability.

Browse the generated 30-Skill Catalog. Every entry links to its canonical SKILL.md and is checked for an independent trigger, input, output, example, validation, fallback, limitation, TIKAZ contribution, and Codex UI metadata.

Every Skill, individually installable

Each link below opens that Skill's own executable contract. Install an orchestrator for the complete routed workflow, or copy only a specialist folder when that focused output is all you need.

Install a complete suite separately

The monorepo is the single source of truth. Seven focused repositories are automatically generated from it so each suite can be discovered, starred, cloned, and installed on its own without creating eight drifting codebases.

Each distribution records the exact canonical commit in DISTRIBUTION.yml, validates independently, and synchronizes weekly or on demand. Cross-suite development and source-of-truth Issues stay in this repository.

📦 Start in 60 seconds

1. Clone

git clone https://github.com/TIKAZI/TIKAZ-AI-Skills.git
cd TIKAZ-AI-Skills

2. Install one Skill

Copy the suite or child folder into the Skill directory supported by your Codex environment. Keep the folder name identical to the name in SKILL.md.

Project-local example on PowerShell:

Copy-Item -Recurse `
  -LiteralPath '.\suites\frontend-design\frontend-design-studio' `
  -Destination '.\.agents\skills\frontend-design-studio'

3. Invoke it naturally

Use frontend-design-studio to turn this bland landing page into one coherent
visual world. Show the first viewport and one representative section before
expanding the whole page.

See copy-ready prompts for every suite.

Method demonstrations: Context Economy fixture · frontend art-direction proof · video evidence card

Context Economy evidence snapshot

Context Economy is positioned as a fidelity-first multimodal context compiler, not a universal “Token compressor.” It converts supported documents into reusable Markdown, preserves facts and source anchors, routes informative visuals and complex tables separately, and exposes uncertain inputs instead of silently flattening them. The checked-in public benchmark separates efficiency from fidelity instead of publishing one composite score:

In practical use, users can paste text or attach files without choosing a pipeline first. Plain text is compacted conservatively; supported documents are converted to Markdown and checked; documents containing images or complex tables split into compact text, visual filtering, bounded vision work, and source fallback; long conversations become recoverable checkpoints.

• long-context profile: 69.7% estimated reduction (4,698 → 1,422, 6 task variants over one synthetic source);
• prompt exact-repeat profile: 14.6% estimated reduction (157 → 134, 4 cases including no-exact-duplicate controls);
• prompt structural-repeat profile: 49.5% estimated reduction (95 → 48, 2 synthetic variants; first wording retained);
• protected facts: 46/46 retained;
• expected anchors: 39/39 retained;
• Text / Hybrid / Source routing: 8/8 labeled cases;
• visual/table filtering: 8/8 labeled cases;
• short-input profile: 143.9% growth, retained publicly to show when the protocol is not economical.

The product claim is the workflow and its verification boundary: build the smallest useful context that remains checkable. The percentages above are scoped evidence, not universal promises. Visual routing is measured; vision-description accuracy is still Pending.

See the generated evidence card, machine-readable metrics, and generated-PDF fidelity evidence. Real-world/scanned PDF fidelity, actual provider Token usage, vision accuracy, and downstream blind-answer quality are still marked Pending.

🔄 Why it is more than a prompt collection

• One owner per deliverable — supporting Skills cannot compete for control.
• Evidence travels with work — claims retain timestamps, confidence, sources, and unresolved gaps.
• Visual proof before scale — frontend, presentation, and illustration workflows verify representative output early.
• Objective completion gates — builds, tests, rendered artifacts, and file state outrank agent confidence.
• Portable by design — no private drive paths, embedded credentials, or universal claims about optional tools.
• Auditable provenance — clean-room work, adapters, references, and removed upstream material are distinguished explicitly.

🗂️ Repository map

TIKAZ-AI-Skills/
├─ suites/                    # seven independently installable workflow suites
│  ├─ context-economy/        # context budgeting, selection, packaging, and recovery
│  ├─ frontend-design/        # flagship: art direction + implementation gates
│  ├─ video-intelligence/     # flagship: evidence-based video understanding
│  ├─ engineering/
│  ├─ knowledge-research/
│  ├─ presentation/
│  └─ visual-content/
├─ examples/                  # copy-ready invocation prompts
├─ scripts/                   # repository validation gate
├─ SOURCES.yml                # source mode, license, and TIKAZ contribution
└─ THIRD_PARTY_NOTICES.md     # research acknowledgements and boundaries

✅ Validation

Every push runs the repository policy gate on GitHub Actions.

powershell -NoProfile -ExecutionPolicy Bypass `
  -File '.\scripts\validate_skills.ps1'

It checks all 30 Skills for structure, attribution, source policy, portability, routing, UI metadata, generated files, and Python syntax. A green check proves repository consistency; optional platform access still depends on the user's environment and permissions.

⚖️ Authorship, sources, and license

The collection architecture, TIKAZ Edition workflows, routing contracts, lifecycle gates, templates, portability rules, and validation scripts are designed, integrated, refactored, and continuously maintained by TIKAZ.

Research references do not transfer authorship. Their URLs, observed licenses, distribution status, and the concrete TIKAZ contribution are recorded in SOURCES.yml and THIRD_PARTY_NOTICES.md.

TIKAZ-authored files are released under the MIT License. Contributions should improve behavior, evidence, portability, or verification—not merely rename an existing Skill. See CONTRIBUTING.md.

🌐 Explore TIKAZ projects

• ⚡ Context Economy — fidelity-first multimodal context compilation with reproducible evidence.
• 🎨 Frontend Design — art-directed frontend workflows from visual proof to implementation.
• 🎬 Video Intelligence — evidence-graded cross-platform video reading and synthesis.
• 🛠️ Engineering — specification-to-release engineering delivery.
• 🔬 Knowledge & Research — traceable research, decisions, and knowledge feedback.
• 📽️ Presentation — narrative-first HTML and editable PowerPoint production.
• 🖼️ Visual Content — illustration, concise writing, lawful music, and publishing QA.