mvanhorn/last30days-skill

This is worth a look if you routinely brief yourself on fast-moving people, products or communities and want the agent to pull the material itself. It is not a replacement for careful research, and its output quality depends on how well the agent can locate the right accounts and threads.

The problem

Suppose you want to know what the community thinks about a person, a product launch or a technology before a meeting. Today that usually means opening Reddit, X, YouTube, Hacker News and a search engine in separate tabs, scanning titles, and guessing which comments and videos actually matter. Then you have to remember which link supported which claim.

The skill changes that workflow: you give the agent a topic, it works out the relevant handles, repositories, subreddits and channels, searches the sources, and returns one synthesized brief with citations project README. The work that moves is the collection and merging step; you still read and judge the resulting brief.

What's different

The main mechanism is depth rather than link lists. The README says results include full YouTube transcripts, top Reddit comments with upvote counts, TikTok captions and Polymarket odds, with items about the same story merged into one cluster and ranked by engagement, relevance and freshness project README. That is a meaningfully different output from a list of search results, because the material is already extracted and grouped before the model writes.

The trade-off is that the initial step is agent-driven resolution of who matters: the skill has to guess the right X handles, subreddits and channels for an ambiguous name. The README presents this as a strength with examples, but any resolver can pick the wrong figure or miss a niche community. Because engagement data is a ranking signal, the brief also inherits the distortions of upvote and like counts—popular takes rise, and quiet expert commentary can stay buried. The final stage is synthesis by a language model, so citations need checking rather than assuming the prose is a faithful summary. The README's claim that the output is "grounded in specific data" is the author's description, not independent verification.

Who should try it

Good fit: someone who uses Claude Code or another Agent Skills host daily and wants a repeatable pre-meeting or pre-build scan of what a community is saying. The install paths are built around those hosts, so friction is low if you already work there.

Poor fit: anyone who needs audited, reproducible research or an offline workflow. The skill depends on live access to several platforms, and the web version requires enabling code execution and file creation in claude.ai before skills will run project README. If you do not already use an agent that supports skills, installing one just for this is a larger commitment than the brief itself warrants.

Requirements and limitations

The most important requirement is an Agent Skills-capable host. The recommended path is the Claude Code marketplace, and the README also documents Grok, claude.ai upload, Claude Desktop, OpenClaw, and a generic installer for 50+ hosts through the Agent Skills CLI install instructions. Pick one method per machine: the README warns that Claude Code will show duplicate /last30days entries if both the marketplace plugin and the npx skills copy are active.

Ongoing upkeep matters. The README distinguishes the source tree from your installed version and points to the changelog and release page, so installs made by re-uploading a file to claude.ai need manual replacement, while marketplace installs are versioned and can be updated in place changelog. Source coverage and account access are also part of the boundary: the brief is only as broad as the sources the skill can actually query, and the README's source table is a feature list, not a guarantee that every platform works from every host.

HotRepo's take

HotRepo first recorded this repository on the daily board on 2026-07-06 at roughly 49.6k stars and last saw it there on 2026-10-26 at 65,522 stars, with 14 observed board days in between; those snapshots do not show continuous growth. The record is too sparse to judge whether adoption persists, so treat the star movement as interest, not proof of maintenance or quality.

Before committing, verify one thing: whether the host you actually use is supported by a documented install path and whether that path handles updates for you. Start with the project README and its install table.