Newsjack.sh: open skills that turn a coding agent into a PR team

The problem Newsjack solves: PR work that never gets done

Most small companies do not lack a story. They lack the routine. Someone has to watch the beat, notice a fresh story the company has standing to comment on, draft an angle before the wave breaks, check the claims, find the right reporter, and send one email instead of fifty. That is a daily loop, and it is the first thing dropped when the week gets busy.

Newsjack's answer is to hand that loop to the agent you already have open. The repository describes the project as "open skills" that make an agent into a PR team, and the README names the intended users directly: founders doing their own PR because an agency quote was too high, agencies running more accounts than people can babysit, marketers at small companies, and anyone whose agent already runs their day.

The framing matters. This is not a SaaS dashboard with seats. It is a set of instruction files your agent reads, plus a CLI, with a bias toward local execution. The trade-off is that you get no web console and no shared team workspace. What you get is a workflow you can open in a text editor and change.

How the skills are organised: detect, act, strategize, AI visibility

The README sorts capabilities into four lanes. Detect covers monitoring and search: /newsjack-monitor-setup builds a profile from your standing, beat topics, competitors, proof assets, spokespeople and feeds; /newsjack-detector looks for stories you can jump on before the wave breaks; /news-search returns dated, attributed articles; /story-origin-check asks who broke a story and how much oxygen is left; /relevance-coarse-filter is described as a cheap high-recall first pass; /newsjack-triage routes surviving stories by whether you actually have standing; and /coverage-tracker watches declared keywords with LLM filtering.

Act covers output: /angle-generator, /headline-generator, /meanest-editor, /crisis-holding, /reactive-comment, /fact-check, /journalist-fit-check, /same-outlet-ranker, /voice-extractor, /find-journalists, and /press-clip. Strategize covers /pr-strategist, /pr-calendar and /newsworthiness-check. The fourth lane, AI visibility, is a chain of eight skills that starts with /build-ai-visibility-panel and ends with /ai-visibility-panel-design.

The invocation model is the interesting part. Skills are called by name, as slash commands, or the agent picks one from a plain description of what you want. That means the quality of your result depends on the agent's routing as much as on the skill text. The pipeline design is deliberate: a coarse filter runs before expensive steps, and triage runs before drafting. The AI visibility lane is the most opinionated, and the README is explicit that /ai-visibility-panel-design gates on a human rather than inventing an "AI visibility score". That refusal is a design choice worth noting, because it is the opposite of what most measurement tools do.

Installing Newsjack and running a first newsjack monitoring setup

The README gives a single install command, a curl piped into bash. The repository also contains an install.sh at the top level, which is what that command is fetching.

curl -fsSL newsjack.sh | bash

After that, the README points humans at a copy-paste prompt for any AI assistant, which asks the agent to install from the GitHub repo and set up daily monitoring for a company. If you are driving the agent yourself, the more direct route is to invoke the setup skill by name, since /newsjack-monitor-setup is documented as the one-time step that captures your standing, beat topics, competitors, proof assets, spokespeople and feeds.

/newsjack-monitor-setup

The README marks this skill as local agent only, so expect it to write state rather than run in a browser chat. That state lives in a SQLite file, and the .env.example documents an optional override for its location:

NEWSJACK_STORE=

With no value set, the file notes the default is ~/.local/share/newsjack/monitor.db. Live news search and journalist enrichment are gated behind a Medialyst key, and the .env.example says that without it, news search falls back to the host's web search and journalist lists stay local. There is also a preferred login path:

newsjack login

Optional X coverage is configured with a bearer token, and the .env.example notes aliases exist for compatibility:

X_BEARER_TOKEN=

A sensible first run is: install, run /newsjack-monitor-setup, then /newsjack-detector, and read the report before wiring anything to Slack. The README says finished reports can optionally be delivered to Slack, but that delivery is tied to the local agent.

Where Newsjack breaks down: platform gaps and local-only skills

The compatibility table in the README is the most useful page in the repository, and also the most sobering. Rows are marked with a check, a warning triangle, or a wrench. ChatGPT is marked with a warning across the skills shown. Several rows, including pr-calendar and the monitoring setup skills, are marked with a wrench rather than a check, and the README states plainly that a few skills reach for a live news index, journalist enrichment, media list research, or locally-saved monitoring state, and that those work best in a local agent.

So the promise "install once, your agent becomes a PR team" is conditional. If your agent is a browser chat, you get a subset. The monitoring profile, the coverage tracker, and the press-clip generator are all flagged as local agent only. If your working style is a chat tab and nothing else, the core detect loop is exactly the part you lose.

The second limitation is data. Journalist lists stay local without a Medialyst key, which means /find-journalists and /journalist-fit-check are working from whatever the agent can see rather than an enrichment service. News search falls back to host web search. That is a real functional difference, not a cosmetic one, and the README does not claim otherwise.

The third is that skills are instructions, not guarantees. A skill like /fact-check extracts claims and flags the shaky ones, but the verification runs through your model. Nothing in the repository suggests an independent fact database. Treat the output as a checklist that a human still signs off on.

How Newsjack differs from a media database or a PR platform

The obvious comparison is a media database and outreach product such as Muck Rack or Cision, and the difference is architectural. Those tools own the data: a maintained journalist and outlet database, contact records, and a hosted interface where a team works. Newsjack owns none of that. It owns the procedure. The journalist side is a skill, /journalist-fit-check, that asks whether a specific reporter would care, and /same-outlet-ranker, which ranks colleagues at one publication so you pitch one person instead of the masthead. Both are reasoning steps over whatever data the agent can reach.

That inverts the cost model. A media database charges for access and gives you a shared workspace. Newsjack charges you in setup and in model calls, and gives you a folder of Markdown you can fork. The README's own framing supports this: it lists PR agencies running more accounts than humans can babysit as a target user, which is a workflow problem, not a data problem.

The second comparison is the generic agent skill ecosystem. Most skill packs are thin wrappers around a single prompt. Newsjack's AI visibility lane is the opposite: a chain of eight skills where /build-ai-visibility-panel drives the six below it in order, /prompt-set-qa gates on provenance and answer leakage, and the panel design keeps aided and unaided denominators separate. Whether or not you care about AI answer visibility, that chain is the clearest statement of the project's intent. It is building a research method, not a prompt library.

Licence, maintenance and the cost of upgrading

Newsjack is MIT licensed, and the repository carries a TRADEMARK.md alongside the LICENSE. MIT covers the code and the skill text, so forking, editing and internal reuse are permitted by the licence text. The trademark file is the boundary to read if you plan to redistribute under the Newsjack name; that is a naming question, not a copyright one. Nothing here is legal advice, and if you are packaging this into a commercial product, the trademark file is the document to put in front of a lawyer.

The release cadence is visible in the tags: v0.1.16 on 2026-08-05, v0.1.17 on 2026-09-14, and v0.1.18 on 2026-09-15. The last push to the repository was on 2026-09-15, and the repository is not archived. Pre-1.0 version numbers mean skill names, file layout and configuration keys can still move between releases, and the CHANGELOG.md at the top level is where that movement is recorded. Upgrade cost is therefore mostly re-reading your local edits: if you have customised a skill file, a pull will conflict with it, and the fix is a merge rather than a config change.

The dependency surface is small. package.json lists a single dependency, playwright-core, which is consistent with the press-clip skill turning a live article URL into a PDF. The CLI itself is Go. Running the local skills means running those dependencies on your machine.

Editorial conclusion

Adopt Newsjack if you already drive an agent daily and want a repeatable PR workflow without hiring: the skills are plain Markdown you can read and edit, and the monitoring state stays on your own machine. Do not adopt it if you need a hosted dashboard, a guaranteed media database, or a single tool that works identically in a browser chat. Before committing, verify three things: whether your target platform supports the skills you actually need (the README's compatibility table marks several rows with a wrench rather than a check), whether you have a Medialyst API key for live news search or are content with host web search, and whether you are willing to run the local agent that the setup skills require.

Frequently asked questions

Official sources

1. elvisun/newsjack on GitHub
2. License: MIT
3. Project website
4. README
5. Releases