news-from-discussions
Skill1 starsChanged 7 months agoMITView on GitHub
Personalized news digest tailored from your past conversations, memory, and projects — plus a checklist of what's pending across your work
In this repository2 other files it gives its agents.
- deep-diveSkill
- secret-guardSkill
The file
Show its rules and text
We split this skill into the rules it gives an agent
14 rules. About 0 are common, 9 are distinctive, and 5 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈9
Distinctive · estimate
This author's own choices. Worth a look.
5
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 19
Read a global TODO file if one exists at ~/.claude/TODO.md or ~/TODO.md.
- Distinctive: line 31
Run ~6 WebSearch queries in parallel:
- Very specific: line 39
Update ~/.claude/news-from-discussions/recent-headlines.md — append today's headlines (date + one-line summary).
- Distinctive: line 45
If no items map to actionable TODOs, skip this section entirely.
- Very specific: line 46
Optional: write to file: If invoked with --out <path> or the user has set the env var NEWS_FROM_DISCUSSIONS_OUT, also write the full …
- Distinctive: line 111
Keep item text short — one line each, no descriptions
- Distinctive: line 118
Recency gate: Only include articles published within the last 2 weeks.
- Very specific: line 119
Query freshness: Use time-specific phrasing covering the full 2-week recency window (e.g.
- Very specific: line 123
Check recent-headlines.md and skip anything already shown in the past week
- Distinctive: line 124
If WebSearch or context files error, skip that section silently — don't break the briefing
- Distinctive: line 128
Only suggest actions when a news item maps to a concrete next step for an existing project — not vague "look into this" items
- Distinctive: line 132
If no items are actionable, skip the section entirely — don't force it
- Distinctive: line 136
NEVER attribute a story to a source unless you verified the story actually appears at that URL
- Distinctive: line 137
If a search result mentions a topic but the URL is a general roundup, find the actual source or label it honestly
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
News From Discussions
A morning briefing that reads who you are from your own files — past conversations, memory, project TODOs — and delivers a news digest tailored to what you've actually been thinking about, not topic tags you picked once and forgot.
Two phases:
- Checklists — pending TODOs across your projects, printed immediately.
- News digest — personalized "For You" + general categories, derived from your memory + project context.
Behavior
Phase 1 — Checklists (print immediately)
- Read a global TODO file if one exists at
~/.claude/TODO.mdor~/TODO.md. Treat it as one project ("global"). - Auto-discover project TODOs: scan
~/for direct subdirectories containing aTODO.md(skip hidden dirs and common noise likenode_modules,Library,Applications). Each becomes a project section. - Present each project as a checklist with pending items first, then completed. Right-align
— #Nnumbers within each section so they line up. - End Phase 1 with: "What do you want to work on?"
The user can reference items by project + number, e.g. "global #2" or "myproject #1".
Phase 2 — News Digest (after Phase 1 is fully printed)
- Gather context — read these (skip any that are missing):
- Memory files at
~/.claude/projects/*/memory/*.md ~/.claude/news-from-discussions/recent-headlines.md(used to avoid repeats across days)
- Memory files at
- Run ~6 WebSearch queries in parallel:
- 2 personalized "For You" — derive queries from the user's interests, recent learnings, current domains, and project context inferred from memory + TODOs. Reflect who they are as a person, not literal TODO text.
- 1 general news headlines
- 1 AI & tech news
- 1 niche of strongest interest — pick the topic that shows up most in their memory/projects (e.g. design, biotech, finance, gaming, cooking, parenting). One category, varies per user.
- 1 community buzz — viral/trending discussions from Hacker News, Reddit, X. Surfaces things that haven't hit mainstream news yet.
- Curate 3–5 items per category. Deduplicate across sections. Skip any headline already in
recent-headlines.md. - Print the For You and News Digest sections (format below).
- Update
~/.claude/news-from-discussions/recent-headlines.md— append today's headlines (date + one-line summary). Trim entries older than 7 days. - Suggested actions — Review today's digest. For each item, ask: does this map to a concrete, actionable next step for one of the user's existing projects? If yes, draft a TODO and assign it to the matching project. Print the Suggested TODOs section and wait for approval. The user can:
- "approve all" — write all suggestions to their respective project
TODO.mdfiles - "approve [project] #N" — approve specific items
- "skip" or "skip [project] #N" — discard
- "edit #N to say X" — modify before writing
- If no items map to actionable TODOs, skip this section entirely.
- "approve all" — write all suggestions to their respective project
- Optional: write to file. If invoked with
--out <path>or the user has set the env varNEWS_FROM_DISCUSSIONS_OUT, also write the full briefing as markdown to that path. This enables downstream delivery (email, Slack, etc.) — see README.
Format
Phase 1
## News From Discussions
**global**
[ ] First pending item — #1
[ ] Second pending item — #2
[x] Completed item — #3
**myproject**
[ ] Pending item — #1
[x] Completed item — #2End with: "What do you want to work on?"
Phase 2
Print after Phase 1 is complete. Separate with a horizontal rule.
---
## For You
Picked based on your interests, recent thinking, and current projects.
1. Headline — [Source](url) · why it's relevant
2. Headline — [Source](url) · why it's relevant
3. Headline — [Source](url) · why it's relevant
## News Digest
**General**
4. Headline — [Source](url)
5. Headline — [Source](url)
6. Headline — [Source](url)
**AI & Tech**
7. Headline — [Source](url)
8. Headline — [Source](url)
**[Niche]** (e.g. Design, Biotech, Cooking — picked from user's signals)
9. Headline — [Source](url)
10. Headline — [Source](url)
**Trending in Tech & Communities**
11. Topic — [Source](url)
12. Topic — [Source](url)
## Suggested Actions
From today's digest — approve, edit, or skip.
**myproject**
- [ ] Concrete actionable item ([source](url)) — #1Rules
Checklists
[ ]for pending,[x]for completed — no bullet prefix, just the brackets- Right-align
— #Nnumbers using spaces so they line up within each project section - Keep item text short — one line each, no descriptions
- Skip projects with no
TODO.md
News Digest
- Phase 2 runs entirely after Phase 1 output — never delay the checklists
- "For You" queries should reflect the user as a person (interests, expertise, recent learnings) — not just search for TODO item text
- The niche category is whatever topic dominates the user's memory/projects. If signals are sparse, default to "Culture & Ideas" with broadly interesting items.
- Recency gate: Only include articles published within the last 2 weeks. Check the actual publication date in the article — listicles with a year in the title are not recent just because they mention the current year.
- Query freshness: Use time-specific phrasing covering the full 2-week recency window (e.g.
"past two weeks"or an explicit date range), not just the current few days. Catches items that went viral 8–14 days ago. - 3–5 items per category, one line each; headlines ≤ ~80 chars
- Link to original articles, not search result pages
- Deduplicate across For You and News Digest
- Check
recent-headlines.mdand skip anything already shown in the past week - If WebSearch or context files error, skip that section silently — don't break the briefing
- Cap total WebSearch calls at 6 to keep latency reasonable
Suggested Actions
- Only suggest actions when a news item maps to a concrete next step for an existing project — not vague "look into this" items
- Each suggestion includes a source link and frames why the item is actionable
- Make suggestions specific: name the niche/tool/data point, connect to the right project
- Group by project, number items within each project for easy reference
- If no items are actionable, skip the section entirely — don't force it
- Approved items get appended to the target project's
TODO.md
Source Attribution
- NEVER attribute a story to a source unless you verified the story actually appears at that URL
- If a search result mentions a topic but the URL is a general roundup, find the actual source or label it honestly
How personalization works
The "For You" picks come from reading the user's own files — memory entries, project TODOs, anything you can find at ~/.claude/projects/*/memory/. The skill doesn't ask the user to list their interests because the interests are already in those files: who they are, what they're building, what they've learned recently, what's confused them. Treat memory as the answer to "what would matter to this person today?"
This is the differentiator: news from your discussions, not from your subscriptions.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

