Skip to content

v0.7.0 — Hot Radar & ai-radar Skill

Latest
Latest

Choose a tag to compare

View all tags
@LearnPrompt LearnPrompt released this 11 Jun 08:20
· 55 commits to master since this release
v0.7.0
d7886b4
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

v0.6 merged scattered messages into story lines. v0.7 answers the next question: with this many stories, what is hot right now? — and ships the first consumer-side skill.

New: ai-radar consumer skill

  • skills/radar/ — install without forking (npx skills add LearnPrompt/ai-news-radar -s ai-radar -g), ask your agent "今天AI圈有什么", get a Chinese brief read directly from this site's public JSON. Zero API, zero key, zero server; the whole data pipeline is forkable.
  • Reproducible demo GIF (vhs tape + script included); two-skill split documented: radar reads, scout selects.

Bole picks v2 (伯乐精选)

  • Hot view: ranks story clusters by multi-source mass × time decay; hides itself when there is no real multi-source heat. Timeline view unchanged.
  • 宁缺毋滥 quality gate: brief slots are earned by multi-source confirmation or score ≥ 0.72; quiet days collapse the block entirely — no empty shell, no padding.
  • Same-cluster suppression: an event reposted outside the 6h merge window now takes one slot instead of three; per-source decay stops one prolific source from filling the brief (live data: 15/20 → 5/20 max share).

Scoring accuracy

  • URL-host-only matching: Google News base64 paths randomly containing llm/gpt no longer turn world news into "model_release" items — a 14-day replay shows 327 false-AI items dropped, 4 A.I.-styled titles recovered, zero genuine losses.
  • Near-duplicate suppression: rewritten syndication (法案 vs 立法) collapses to the more authoritative copy.

New tool: scoring backtest

  • python scripts/backtest_scoring.py --days 14 --baseline-rev <rev> replays the archive through two scorer versions and reports flips, label moves, and per-site deltas. House rule going forward: scoring changes ship with a ≥14-day replay report.

UI & performance

  • Per-group render cap with "展开剩余 N 条" (first paint: 806 → 523 cards, page height −30%)
  • Mobile: stats stay in two compact columns — content is one swipe away instead of three screens

Reliability

  • Fixed: daily-brief.json / stories-merged.json had been silently frozen for 8 days (workflow git-add whitelist omission); commit step hardened with git add data/ + a loud unstaged-files check
  • Added MIT LICENSE (README had claimed it all along)

Tests: 80 passing (24 added this release).

🤖 Generated with Claude Code

Assets 2
Loading