daily-brief
Builds the daily news brief: today's stories, weather and markets on one phone-first page. Use it when the user asks for their brief, or a routine opens one.
Install
npx skills add https://github.com/cgoinglove/thursday-agent/tree/main/skills/daily-brief
claude plugin marketplace add https://llmmart.ai/marketplace.json && claude plugin install cgoinglove-thursday-agent@llmmart
git clone https://github.com/cgoinglove/thursday-agent.git
The skills CLI installs just this skill, for any of its supported agents. Claude Code installs the whole cgoinglove/thursday-agent collection as a plugin from our marketplace. Git is the plain clone.
Skill manifest
Daily brief
Four scripts do everything but choosing and writing. They sit in this skill's scripts/ folder and
node runs them; nothing to install, no key. <scratch> is this job's own folder under scratch/:
write it out in full, and set S= again in every bash call — each call is a new shell.
S=<this skill's folder>/scripts
# Gather, one call: every topic as "<label>=<query>", in the edition the user reads, with the
# glance beside it when they want one
node $S/news.mjs "AI=AI OR OpenAI OR Anthropic" "Startups=startup funding" \
--lang en --country US --hours 30 --out <scratch>/cand.json [--avoid a.com] [--prefer b.com]
node $S/glance.mjs --weather "<their city>" \
--markets "S&P 500:SP500,EUR/USD,Bitcoin:CBBTCUSD" --out <scratch>/glance.json
# Read, one call: the stories you chose, from their publishers, photos saved
node $S/story.mjs <scratch>/cand.json a1 a4 b2 c1 c3 --out <scratch>/stories
# Their words, one call a file: story.mjs prints the `cat` line for each
cat <scratch>/stories/text-1.md
# Lay out, one call: after you have written <scratch>/brief.json yourself
node $S/page.mjs <scratch>/brief.json --look <scratch>/look
# A PDF beside it, only when they ask for one: a brief sent to their phone goes with its PDF
node $THURSDAY_SKILLS/artifact/scripts/document.mjs pdf <the page's path>
- The ids are all you pass between steps.
news.mjsprints an id per story (a1,b3);story.mjstakes those ids;brief.jsonnames the same ids. A url, a photo path or a number is never typed: the page takes them from the scripts' files. - What the scripts print is what you read — except the stories' own words.
cand.jsonandstories.jsonare for the next script; opening one costs tens of thousands of tokens and says nothing the printout does not. The articles' text is the one thing too long to print:story.mjswrites it intotext-<n>.mdfiles that each fit one read, and prints thecatline for each. Those files are the only source for a summary. Choose by reading the list yourself — never with a script that scores it. - A story
news.mjsdid not print is not in the brief. The very article a brief of the last four days carried is left out before you see it; the same news in another article is yours to judge, against the headlines those briefs told, printed under the list. A storystory.mjsprints asNO TEXTwas refused by its publisher: drop it for one of your spares; withNO PICTURE, keep it or swap it. - A brief is remade, not added to. Run the whole thing again the same day and today's own page
does not count as already told, so the same stories come up again and
page.mjswrites over it. - The browser opens itself where a step needs it. A Google News link does not hold the
publisher's address, so
story.mjsfollows the chosen stories' links in this job's browser;--lookrenders the page in a headless one of its own. Nothing else here needs one.
references/brief.md is the format: what brief.json holds, how to choose, how to write a
summary and the spoken version. references/preferences.md is what to keep about the user and
the one question a first brief asks.
Files (thursday-agent)
-
page
-
brief.css 10.3 KB · in bundle
-
-
references
-
brief.md 8.3 KB
Loading… -
preferences.md 3.1 KB
Loading…
-
-
scripts
-
glance.mjs 6.8 KB · in bundle
-
lib.mjs 4.9 KB · in bundle
-
news.mjs 10.1 KB · in bundle
-
page.mjs 20.1 KB · in bundle
-
story.mjs 11.9 KB · in bundle
-
-
SKILL.md 3.3 KB
--- name: daily-brief description: "Builds the daily news brief: today's stories, weather and markets on one phone-first page. Use it when the user asks for their brief, or a routine opens one." --- # Daily brief Four scripts do everything but choosing and writing. They sit in this skill's `scripts/` folder and `node` runs them; nothing to install, no key. `<scratch>` is this job's own folder under `scratch/`: write it out in full, and set `S=` again in every bash call — each call is a new shell. ```bash S=<this skill's folder>/scripts # Gather, one call: every topic as "<label>=<query>", in the edition the user reads, with the # glance beside it when they want one node $S/news.mjs "AI=AI OR OpenAI OR Anthropic" "Startups=startup funding" \ --lang en --country US --hours 30 --out <scratch>/cand.json [--avoid a.com] [--prefer b.com] node $S/glance.mjs --weather "<their city>" \ --markets "S&P 500:SP500,EUR/USD,Bitcoin:CBBTCUSD" --out <scratch>/glance.json # Read, one call: the stories you chose, from their publishers, photos saved node $S/story.mjs <scratch>/cand.json a1 a4 b2 c1 c3 --out <scratch>/stories # Their words, one call a file: story.mjs prints the `cat` line for each cat <scratch>/stories/text-1.md # Lay out, one call: after you have written <scratch>/brief.json yourself node $S/page.mjs <scratch>/brief.json --look <scratch>/look # A PDF beside it, only when they ask for one: a brief sent to their phone goes with its PDF node $THURSDAY_SKILLS/artifact/scripts/document.mjs pdf <the page's path> ``` - **The ids are all you pass between steps.** `news.mjs` prints an id per story (`a1`, `b3`); `story.mjs` takes those ids; `brief.json` names the same ids. A url, a photo path or a number is never typed: the page takes them from the scripts' files. - **What the scripts print is what you read — except the stories' own words.** `cand.json` and `stories.json` are for the next script; opening one costs tens of thousands of tokens and says nothing the printout does not. The articles' text is the one thing too long to print: `story.mjs` writes it into `text-<n>.md` files that each fit one read, and prints the `cat` line for each. Those files are the only source for a summary. Choose by reading the list yourself — never with a script that scores it. - **A story `news.mjs` did not print is not in the brief.** The very article a brief of the last four days carried is left out before you see it; the same news in another article is yours to judge, against the headlines those briefs told, printed under the list. A story `story.mjs` prints as `NO TEXT` was refused by its publisher: drop it for one of your spares; with `NO PICTURE`, keep it or swap it. - **A brief is remade, not added to.** Run the whole thing again the same day and today's own page does not count as already told, so the same stories come up again and `page.mjs` writes over it. - **The browser opens itself where a step needs it.** A Google News link does not hold the publisher's address, so `story.mjs` follows the chosen stories' links in this job's browser; `--look` renders the page in a headless one of its own. Nothing else here needs one. `references/brief.md` is the format: what `brief.json` holds, how to choose, how to write a summary and the spoken version. `references/preferences.md` is what to keep about the user and the one question a first brief asks.
Comments (0)
Sign in to join the conversation.
Reviews (0)
No reviews yet.
No comments yet.