AI Daily Card

AI Daily Card (Beta)

Create the two fixed deliverables for the current Asia/Shanghai day:

• output/AI日报-今日.md
• output/AI日报-今日.png

The project containing this file is the canonical Skill root. Resolve paths from that root even when the Skill was discovered through a symlink.

Idempotency gate

Before any network request or write, read only the Markdown frontmatter. If date equals today in Asia/Shanghai and status is success, stop silently. Do not invoke the collector, renderer, package manager, or any network tool.

If the file is missing, stale, or not successful, run the complete workflow.

Complete workflow

1. Ensure dependencies were installed with npm ci and npx playwright install chromium. Do not silently install during a Scheduled run; report the missing dependency.

2. Collect the previous 24 hours into .runtime/raw.json:

   python3 scripts/fetch_aihot.py \
     --state-dir .runtime \
     --output .runtime/raw.json
   

   A cache hit is allowed. An empty result, missing field, or interrupted page is a hard failure.

3. Read .runtime/raw.json and curate by judgment, not score alone: deduplicate the same event, choose two headlines, then target 15–18 unique items across the configured sections. A weak section may be empty or omitted.

4. Write .runtime/curated.json using the curated contract. Follow these evidence rules:

   • Keep each selected item's display_title exactly equal to one source title.
   • Build note only by joining one or two exact, contiguous excerpts from the source summaries; list the same excerpts in note_evidence.
   • Never add text for an empty summary.
   • Preserve the canonical source item's exact url, source, publishedAt, and score; include every merged source in source_urls.
   • Do not repeat a headline in a section. Do not invent a category, fact, or number. If traceability is uncertain, omit the item.

5. Render and validate. The renderer verifies the raw SHA, exact source fields, evidence excerpts, every displayed numeric token, unique counts, contrast, required Noto Sans SC font, 750px CSS width, 1500px PNG width, and horizontal overflow before replacing fixed outputs:

   node scripts/render_card.mjs \
     --input .runtime/curated.json \
     --template templates/card.html \
     --state-dir .runtime \
     --output-dir output
   

6. Report both fixed paths and the renderer result. Leave human review and group sending to the user.

Curation shape

• Default sections: 模型与产品, 芯片与算力, 商业与资本, 研究与观点, 机器人与硬件.
• The target is two headlines plus 13–16 section items, 15–18 total.
• Merge duplicate coverage by keeping one displayed item and attaching all relevant source_urls; never combine unsupported claims.

Customization

Set brand.name, theme.accent, theme.background, and sections[].name in the curated JSON. Contrast checks still apply. To replace AI Hot, provide a JSON source conforming to the input contract and keep equivalent raw provenance.

Boundaries and attribution

• Footer attribution must include AI Hot 聚合（aihot.virxact.com）.
• Never use aihot-skill/* as the User-Agent.
• Never read or modify CC daily-startup, the knowledge-incubation pipeline, ~/.claude/skills/, Slack, WeChat, or any group-send automation.
• Do not push, publish, deploy, or create a remote repository.
• This beta Skill is licensed under CC BY 4.0.