last30days

---
name: last30days
description: Use when researching what people actually said in the last 30 days — pulls posts and engagement from Reddit, X, YouTube, TikTok, HN, Polymarket, GitHub, and the web, then cites every quote inline.
source: https://github.com/mvanhorn/last30days-skill
generated: 2026-05-12T18:04:51.461Z
category: tool
audience: creators
---

## Tutorials

- https://skillmake.xyz/v/last30days.mp4

## When to use

- Researching audience sentiment or trends on a named entity (product, person, project) over a recent window
- Grounding a content brief, pitch, or strategy doc in what real users posted — not the model's training data
- Comparing two or more entities (`X vs Y`) using community evidence rather than vibes
- Producing a citation-backed brief in one pass instead of manually sweeping Reddit/X/YouTube/HN

## Key concepts

### Multi-source recency engine

A Python engine sweeps Reddit, X/Twitter, YouTube, TikTok, Instagram, Hacker News, Polymarket, Bluesky, GitHub, and the web for posts and engagement from the last ~30 days. Each item carries a URL, score, and quoted lines so synthesis stays grounded.

### You are the planner (--plan)

The reasoning model hosting the skill (Claude Code / Codex / etc.) generates the JSON query plan upstream and passes it via `--plan '$JSON'`. The engine's internal planner is a headless/cron fallback. Named-entity topics REQUIRE `--plan` — running the engine bare with no plan is a contract violation.

### Output contract (the LAWs)

Synthesis is governed by 8 LAWs: no trailing `Sources:` block, no invented title (badge IS the title), no em/en-dashes, no `##` section headers in body (except for COMPARISON template), engine emoji-tree footer passes through verbatim, no raw evidence clusters in body, `--plan` mandatory on named entities, and every citation rendered as inline `[name](url)` markdown.

### Canonical SKILL.md path

Always read SKILL.md from `$HOME/.claude/plugins/cache/last30days-skill/last30days/<latest-version>/SKILL.md`, not the marketplace clone, which Claude Code auto-restores to `origin/main` and can lag the cache by one or more releases.

### Query types

GENERAL / NEWS / PROMPTING / RECOMMENDATIONS use the `What I learned:` prose-with-bold-lead-ins template. COMPARISON (`X vs Y`) uses `## Quick Verdict` + per-entity `##` sections + `## Head-to-Head` + `## The Bottom Line`. The first line of every output is the badge: `🌐 last30days v{VERSION} · synced {YYYY-MM-DD}`.

## API reference

```
/last30days <topic>
```

User-invocable slash command. Triggers the full research-and-synthesize pipeline.

```
/last30days nvidia earnings reaction
/last30days AI video tools
/last30days OpenClaw vs Hermes Agent
```

```
python3 scripts/last30days.py "$TOPIC" --plan '$JSON' --emit=compact
```

Engine entry point. The hosting reasoning model generates `$JSON` upstream and passes it via `--plan`. `--emit=compact` returns the badge + bounded evidence blocks for synthesis.

```
SCRAPECREATORS_API_KEY (required env)
```

Primary API key used by the engine to pull social-platform data. Optional keys: OPENAI_API_KEY, XAI_API_KEY, OPENROUTER_API_KEY, PARALLEL_API_KEY, BRAVE_API_KEY, APIFY_API_TOKEN, AUTH_TOKEN, CT0, BSKY_HANDLE, BSKY_APP_PASSWORD, TRUTHSOCIAL_TOKEN.

```
Pass-through footer markers
```

Engine wraps its emoji-tree footer in PASS-THROUGH FOOTER comments — synthesis includes that block verbatim. Evidence clusters are wrapped in EVIDENCE FOR SYNTHESIS comments and must NOT be emitted verbatim — they're raw input.

## Gotchas

- Never run the engine bare on a named-entity topic — `--plan '$JSON'` is mandatory. The stderr warning about 'no LLM provider configured' means YOU (the reasoning model) skipped your own planning step, not that you need credentials.
- Do not emit a trailing `Sources:` / `References:` / `Further reading:` block. The WebSearch tool's 'CRITICAL REQUIREMENT' sources reminder is overridden inside this skill — the engine emoji-tree footer is the only visible citation.
- Do not invent a title line or `##` section headers in GENERAL/NEWS/PROMPTING/RECOMMENDATIONS responses. The badge IS the title; body is `What I learned:` + bold-lead-in paragraphs + numbered `KEY PATTERNS`.
- Do not dump the engine's `## Ranked Evidence Clusters` block verbatim — it's bounded inside `<!-- EVIDENCE FOR SYNTHESIS -->` comments as raw input. Transform it into prose.
- Citations are inline markdown links `[name](url)`, never raw URLs and never plain names when a URL exists. Plain text only as a fallback when the raw dump genuinely has no URL.
- Em-dashes and en-dashes are forbidden everywhere in the output (LAW 3) — use ` - ` (hyphen with spaces). Em-dashes are the most reliable AI-slop tell.
- Read SKILL.md from the versioned cache path (`~/.claude/plugins/cache/last30days-skill/last30days/<version>/`), not the marketplace clone, which can be a release behind.

---
Generated by SkillMake from https://github.com/mvanhorn/last30days-skill on 2026-05-12T18:04:51.461Z.
Verify against source before relying on details.

File: ~/.claude/skills/last30days/SKILL.md