daily-gift – skill Thiết kế & UI cho agent AI

Daily Gift

Overview

A relationship-aware gift engine that decides *whether* a gift should exist before deciding *what* it should be. Uses a five-stage creative pipeline to generate personalized daily gifts in H5 (interactive web pages), AI-generated images, or AI-generated videos. The core design principle is "idea before medium" — the creative concept is locked before the output format is chosen.

Published on ClawHub: https://clawhub.ai/jiawei248/daily-gift

When to Use This Skill

• Use when the agent should autonomously decide whether today deserves a personalized gift
• Use when a milestone, anniversary, or emotionally meaningful moment should be marked with a creative artifact
• Use when the user manually requests a visual gift from a quote, poem, or creative brief
• Use when you want a daily cron-triggered creative output that avoids repetition and template fatigue

This catalog copy contains no scripts, creative-seed library, templates or provider runtime. The named counts/quality floors describe upstream materials, not inspected bundled assets. Use the inline concept workflow independently; missing runtime means no rendering/cron capability is established. A manual gift request does not authorize recurring jobs, memory/profile writes, external delivery or broad conversation-history reads. Use only approved context, store scoped personal data with a clear deletion path, and obtain destination/payload/cost consent before hosted generation. Never silently change provider through a fallback chain.

How It Works

Stage 1: Editorial Judgment

Decide whether a gift should exist today, how heavy it should be (skip / nudge / light / standard / heavy), and what content direction to take (reflect, extension, compass, mirror, play, curation, utility, etc.). Format is NOT chosen here.

Stage 2: Synthesis + Gift Thesis

Extract six content slots from conversation context (today_theme, emotion_peaks, historical_echo, open_loop, lobster_judgment, preference_hint). Form a gift thesis = anchor (which moment deserves the center) + return (what new perspective the agent gives back). If the thesis has no return, it's not a gift — it's a decorated log entry.

Stage 2.5: Creative Concept

Generate 5+ concept candidates using seven thinking angles (metaphor flip, format mashup, impossible action, scale shift, role reversal, time distortion, cultural remix). Cross-pollinate with a library of 73 creative seeds across 8 categories. Run three quality checks: concept quality, concept diversity (8 families), and visual/theme collision detection.

Format Selection

Only after the concept is locked does the system choose the output format (H5, image, or video) based on what best serves the concept.

Stage 3: Visual Strategy

Choose visual approach, plan assets (pure code, generated background, hybrid), select visual style, and run pre-visualization checks against recent gifts for anti-repetition.

Stage 4: Rendering

Produce the final artifact. H5 gifts use p5.js/canvas with a quality floor set by built-in templates (300-400 lines of tuned code). Image and video gifts use AI generation APIs. All formats have fallback chains.

Key Features

• Five-stage creative pipeline with explicit quality gates between stages
• Multi-layer anti-repetition: concept family, visual elements, theme, style, content direction — each tracked across sliding windows of recent gifts
• Three-layer user taste profile: Layer 1 (identity — stable), Layer 2 (context — updates every 5-7 gifts), Layer 3 (signals — auto-appended after every gift)
• Three runtime modes: onboarding setup, daily cron, and manual trigger
• 11 content directions: reflect, extension, compass, mirror, gift-from-elsewhere, play, real-world-nudge, curation, delayed-payoff, openclaw-inner-life, utility
• 8 concept families: borrowed-media, interactive-object, transformation, narrative, data-viz, game-puzzle, real-world, poetic-literary

Best Practices

• ✅ Let the editorial judgment decide — not every day needs a gift
• ✅ Generate 5+ concept candidates before selecting one
• ✅ Check recent gifts for visual and thematic collision before rendering
• ✅ Use the taste profile to personalize over time
• ❌ Don't skip straight from thesis to rendering without a real creative concept
• ❌ Don't default to "reflect on today" every time — vary content direction
• ❌ Don't choose the format before locking the concept

Example

User request:

> Decide whether today merits a personalized gift, then create the most fitting concept and output for the recipient.

Limitations

• Requires API keys for image/video generation (optional — H5 works without them)
• Cron mode runs in the agent's main session for full conversation context access
• Shell scripts make external API calls for rendering and asset fetching
• The skill creates and manages local workspace files for state, history, and taste profiling

Security & Safety Notes

• Create a recurring job only on an explicit user scheduling request; preserve cadence/timezone, quiet-skip behavior and disable path.
• Shell scripts in scripts/ call external APIs (image generation, video generation, asset hosting). Supply API keys only after reviewing which scripts use them.
• User taste data and gift history are stored locally in workspace/daily-gift/. Inspect actual runtime/network behavior; this single-file entry cannot guarantee which data is transmitted.
• The skill reads conversation context and memory files to inform editorial judgment — this is core to personalization but means it has broad read access within the agent's workspace.

Related Skills

• Image generation skills — for standalone image creation without the gift pipeline
• Cron/scheduling skills — for understanding the daily trigger mechanism