Link Triage Explainer · X profile + GitHub repo

/last30days — agent-led search over recent social signal

Source URL: https://x.com/slashlast30days

Upstream repo: mvanhorn/last30days-skill

Profile inspected through X API and followed to the public GitHub repo. The profile describes /last30days as “Skill for searching /last30days on X, Reddit and the web” by @mvanhorn; the repo positions it as an Agent Skills package that searches recent public/community sources and synthesizes them into a grounded brief.

Agent Skillresearchsocial searchPython 3.12+Hermes-compatiblemulti-harness

What this link is

Classification

An X profile for a tool, not just a person/post. The important outbound link is the GitHub repository mvanhorn/last30days-skill, which contains the actual Agent Skill and Python research engine.

Core promise

Run a query like /last30days Peter Steinberger and have an agent gather recent signal from Reddit, X, YouTube, TikTok, Hacker News, Polymarket, GitHub, Digg, and web sources, then rank/synthesize what people are actively engaging with.

Visible traction

The X profile had ~2.6K followers and only 11 posts at inspection time. Launch posts claim 20K+ GitHub stars and heavy usage. Treat that as momentum signal, not proof of result quality.

Why it matters

/last30days is trying to solve a real agent gap: most models do not have reliable, fresh, multi-platform “what are people saying right now?” access. Google is SEO/editorial-biased; individual platforms are walled gardens; model training data is stale. This tool packages the messy platform-specific access paths into a skill an agent can invoke.

Question / topic
  → planner chooses source mix
  → source adapters fetch recent evidence
  → engagement/social/money signals rank evidence
  → judge/synthesizer creates a brief
  → raw artifacts are saved for audit/reuse

For Ananth/Dab, this is closest to a reusable “community intelligence probe”: before building, buying, debugging, or briefing, ask what high-signal public communities and markets have surfaced in the last month.

Stack and architecture

Skill surface: skills/last30days/SKILL.md defines the agent-facing contract. The repo emphasizes that the product is the slash-command skill, not the direct CLI.

Engine: skills/last30days/scripts/last30days.py plus scripts/lib/ modules implement source querying, entity extraction, ranking, rendering, and storage.

Source adapters: modules cover Reddit public JSON, Hacker News, Polymarket, GitHub, YouTube/yt-dlp, X via browser cookies/xAI/ScrapeCreators, TikTok/Instagram/Threads/Pinterest/Bluesky/TruthSocial, Digg, Perplexity, and generic web backends.

Storage/output: default raw outputs land under ~/Documents/Last30Days/; configurable with LAST30DAYS_MEMORY_DIR, --save-dir, and --save-suffix. v3.3.0 added --emit=html for shareable briefs.

Distribution: installable through Claude Code plugin marketplace or npx skills add mvanhorn/last30days-skill -g for Codex, Cursor, Copilot, Gemini CLI, Hermes, and other Agent Skills hosts.

Source coverage and what each source adds

Source	Signal	Operational note
Reddit / HN	Community pain, workarounds, developer consensus, top comments.	Mostly free/public; strong default value.
X / TikTok / Instagram / Threads / Pinterest / Bluesky	Fresh opinions, creators, cultural/market signal, screenshots/video discourse.	Requires cookies or API/vendor access for best coverage; can be brittle.
YouTube	Long-form transcript evidence and quotable moments.	Needs `yt-dlp`; datacenter IPs may hit YouTube bot walls.
Polymarket	Market odds backed by money, useful for “what do people bet will happen?” questions.	Free source; interpret as market signal, not ground truth.
GitHub	Repo velocity, releases, issues, discussions, PR activity.	Uses `gh` auth when available.
Digg / web / Perplexity	Curated trend clusters, citations, broader editorial context.	Some backends need API keys; Digg is auto-enabled when its CLI is present.

Install and use path for Ananth/Dab

Hermes install path

hermes skills install mvanhorn/last30days-skill --force

Repo docs say this deploys to ~/.hermes/skills/research/last30days/. Then invoke in Hermes as last30days "your research topic".

Generic Agent Skills path

npx skills add mvanhorn/last30days-skill -g

Global user install across supported harnesses. The repo frames this as the modern multi-harness path.

Claude Code marketplace path

/plugin marketplace add mvanhorn/last30days-skill
/plugin install last30days

Recommended by the README for Claude Code because the marketplace handles updates.

Zero-config baseline: Reddit, Hacker News, Polymarket, and GitHub work immediately if their public/CLI access is available. YouTube needs yt-dlp. X and richer social/video sources need cookies or API/vendor credentials.

Maturity signals

Strong repo shape: dedicated README, CONFIGURATION, HERMES_SETUP, AGENTS, CONCEPTS, changelog, tests, fixtures, docs/solutions, and release notes.
Real test surface: many tests cover providers, rendering, env/config, pipeline, YouTube, X, Reddit, Polymarket, GitHub, and setup flows.
Rapid iteration: v3.3.0 changelog says ~75 PRs merged in a week plus community fixes.
Multi-harness awareness: docs explicitly warn that features must be integrated into SKILL.md so agents know how to invoke them.
Fragility risk: the highest-value sources are also the least stable: X cookies/API access, YouTube anti-bot behavior, TikTok/Instagram vendors, and API-key quotas.
Quality risk: engagement-weighted evidence is useful but can amplify noisy communities. Use it for signal discovery, not final truth.

Fit for Mission Control

Link Triage expansion

When a captured link points to a tool/topic, Dab could run /last30days to discover recent use cases, objections, competitors, and implementation pain before writing a Read explainer.

Daily Podcast / briefing pipeline

It can act as a source collector for “what moved in AI/tools today?” especially if paired with Digg, X, HN, Reddit, YouTube, and saved raw outputs.

Pre-build validation

Before turning an idea into a project, ask /last30days for complaints, workarounds, repo velocity, and community demand. That is more useful than generic web search.

People/company prep

The README’s meeting example is directly useful: recent tweets, podcasts, repo activity, Reddit/HN mentions, market narratives, and live context before a call.

Recommended next step

Do a low-risk local install/evaluation rather than adopting it as infrastructure immediately:

hermes skills install mvanhorn/last30days-skill --force
# then run one bounded test query:
last30days "Mission Control personal AI assistant workflows" --search=reddit,hn,github --days=30

Evaluate three things only: output quality, runtime/reliability on the Pi, and whether raw output can be saved/linked cleanly into Notice Board artifacts. Do not wire credentials for X/TikTok/Instagram until the free-source baseline proves useful.

Bottom line

/last30days is a high-fit research primitive for Dab: it turns “what are people actually saying/using this month?” into an agent-invokable skill. The opportunity is not replacing web search; it is adding fresh social/community/market evidence to Mission Control workflows. The caveat is operational: the broader the source mix, the more credentials, cookies, quotas, and platform breakage Dab has to manage.

Generated by Dab’s autonomous Link Triage explainer pass on 2026-05-22. Evidence used: X API profile/timeline metadata for @slashlast30days and public GitHub repository docs/code tree for mvanhorn/last30days-skill.