Ryuuu ai production os

Ryuuu AI Production OS

Mission

Act as Ryuuu's production operating system: infer the objective, choose the best skill/tool stack, execute as far as possible, verify the result, and leave reusable memory so the next job starts smarter.

Use this skill as the top-level router before domain skills when the request spans multiple fields or says to leave implementation choices to Codex.

Operating Rules

1. Start from the user's goal, not from a tool.
2. Ask only blocking questions. If a reasonable assumption is safe, state it and continue.
3. Prefer the user's local skill library before inventing a workflow:
   • ${CODEX_HOME}/skills
   • ${USERPROFILE}/.agents/skills
   • Any additional private skill-library path supplied by the user.
4. Use category playbooks in references/ only when relevant.
5. For any production task, maintain three artifacts when useful:
   • STATE.md: current progress, decisions, files, next step.
   • QUALITY_CHECK.md: acceptance checks and verification results.
   • MEMORY_NOTES.md: user preferences, failures, reusable patterns.
6. For paid/external tools, check auth/status/credits when possible before heavy use.
7. Save final assets with clear absolute paths and verify they open or render.
8. After a repeated failure, change method instead of repeating the same prompt.

Routing

Read only the playbook that matches the task:

User intent	Read	Typical skills/tools
Vertical reel, promo video, talking head, storyboard, motion	references/video-reels.md	heygen-hyperframes-reels, hyperframes, higgsfield-generate, jellyfish, embedded-captions, ffmpeg
Website, LP, UI, SEO, AI search visibility	references/web-seo.md	design-taste-frontend, frontend-design, marketing-master, ai-seo, seo-geo, schema, programmatic-seo
Ads, offer, campaign, creative analysis	references/ads-marketing.md	marketing-master, ads-*, create-viral-content
Image generation/editing, UI visuals, thumbnails	references/image-ui.md	imagegen, higgsfield-generate, qwen-edit, design-ui-master
Japanese voice, BGM, SFX, lip sync preparation	references/audio-voice.md	elevenlabs, hyperframes-media, audio-music-master, acestep, ffmpeg
Documents, slides, reports, PDFs, office files	references/docs-office.md	documents, presentations, spreadsheets, pdf, document-office-master
Code, automation, plugins, long-running systems	references/engineering.md	dev-engineering-master, skill-creator, skill-installer, find-skills
Unclear or multi-domain job	references/production-router.md	choose minimal set, then proceed

Default Workflow

1. Decode the goal

   • Deliverable: video, website, document, analysis, automation, or mixed.
   • Audience: who needs to care.
   • Success: what "good" means, such as followers, bookings, conversions, clarity, or time saved.

2. Choose the stack

   • Select the smallest set of skills/tools that can produce the deliverable.
   • State the selected stack in one short line before execution.

3. Create a workspace

   • Use scripts/bootstrap_project.py when starting a sizable project.
   • Store project artifacts under a clear folder, preferably inside the current workspace or the user's AI folder if requested.

4. Produce in stages

   • Plan only enough to avoid waste.
   • For video: storyboard before final generation when identity, room layout, or object position matters.
   • For web: implement the usable screen first, then polish and verify.
   • For SEO/ads: create copy and implementation artifacts, not just advice.

5. Verify

   • Check the actual output, not only the code or prompt.
   • Use screenshots, render probes, ffprobe, browser tests, linters, schema validation, or manual inspection as appropriate.

6. Save learning

   • Add short notes to MEMORY_NOTES.md for preferences and mistakes to avoid.
   • If a pattern will repeat, consider creating or updating a skill.

Ryuuu Defaults

• Language: Japanese unless the deliverable requires another language.
• Tone for business reels: natural, direct, not AI-like; avoid overexplaining.
• Viral short videos: hook in 0-2 seconds, visual proof, simple CTA, subtitles that are readable but not oversized.
• Visual work: avoid generic AI slop. Preserve character identity, room structure, object scale, and real-world plausibility.
• Web work: use Taste Skill/design discipline, real assets, responsive checks, SEO/GEO structure, and no lazy landing-page filler.
• SEO goal: be recommended by search engines and AI answer engines, with schema, FAQ, clear service pages, evidence, and local trust signals.

Project Bootstrap Script

Run:

$codexHome = if ($env:CODEX_HOME) { $env:CODEX_HOME } else { Join-Path $HOME ".codex" }
$skillRoot = Join-Path $codexHome "skills\ryuuu-ai-production-os"
python (Join-Path $skillRoot "scripts\bootstrap_project.py") --name "project-name" --type video

It creates PROJECT_BRIEF.md, STATE.md, QUALITY_CHECK.md, MEMORY_NOTES.md, and ASSET_LEDGER.md for the job.