v2.6.0

The AI world reinvents itself every month. This skill keeps you current.

\`/last30days\` researches your topic across **Reddit, X, YouTube, Hacker News, Polymarket, and the web** from the last 30 days, finds what the community is actually upvoting, sharing, and betting on, and writes you a prompt that works today.

## Headline Features

**1. Hacker News as a 5th source.** Every run now searches HN stories and comments via the Algolia API. No API key, no setup — it just works. Developer and technical topics especially benefit.

**2. Polymarket prediction markets as a 6th source.** Real money on real outcomes. When markets exist for your topic, odds surface as the highest-signal data points in the research. Outcome-aware scoring finds markets where your topic is an *outcome* in a broader market, not just the title — so \`/last30days Arizona Basketball\` finds the NCAA Tournament Winner market even though "Arizona" only appears as an outcome.

**3. X handle resolution.** For people, brands, and products, the skill automatically resolves the official X handle and searches their posts directly — surfacing content they posted that doesn't mention their own name.

**4. Agent-native invocation.** The skill can now be called by other agents via the Skill tool. Pass \`--agent\` for non-interactive report output — skips the intro, clarifying questions, and wait prompts, and returns a complete structured research report the calling agent can use directly.

**5. Cross-source linking.** Items that appear across multiple platforms (Reddit + HN, X + YouTube) are flagged with \`[also on: ...]\` tags and weighted higher in synthesis — cross-platform signals are the strongest evidence.

## What's New Since v2.1.0

### Added
- Hacker News source (Algolia API, free, no auth)
- Polymarket prediction markets source with outcome-aware query expansion
- Polymarket synthesis instructions — odds woven into narrative as supporting evidence
- X handle resolution for person/brand/product topics
- Cross-source linking and YouTube relevance scoring
- \`--agent\` flag for non-interactive agent-mode report output
- OpenAI 429 rate limiting with exponential backoff (thanks [@tjarko](https://github.com/tjarko))

### Fixed
- Zero-result source lines now hidden from stats (no more "0 stories (no results this cycle)")
- HN ordering in stats block
- X search: unfiltered handle search no longer skipped for entity-extracted handles
- Handle resolution broadened to products/tools/brands with parody account detection
- Polymarket: two-pass query expansion for outcome-based market discovery

### Changed
- Stats block format: \`├─ 📊 Polymarket\` with market count + top odds inline
- Synthesis prioritizes cross-platform signals over single-source findings
- Security docs updated: skill is now agent-invocable (inline mode, not forked)

## Upgrading

```bash
git pull
```

Or reinstall to your skill directories and run \`bash scripts/sync.sh\`.

---

*Community contributors: [@tjarko](https://github.com/tjarko) — OpenAI rate limit fix*