Tech Discoveries

Category: Research Example Model: balanced Updated: 2026-05-25

Curate a weekly list of relevant technical links from sources you already trust.

Quick Start

Prerequisites

Source list: newsletters, RSS feeds, GitHub Trending, HN, selected subreddits, or local bookmarks.
Web search/fetch enabled if sources require it.
Delivery channel configured and allowlisted.
Clear interests and skip rules.

Add The Job

openclaw cron add \
  --name "tech-discoveries" \
  --cron "0 8 * * 0" \
  --timezone "America/Los_Angeles" \
  --session isolated \
  --message "Generate weekly tech discoveries for [YOUR_INTERESTS]. Check [YOUR_SOURCES]. Curate 5 to 10 items. For each item include title, link, one-sentence summary, and why it matters to my interests. Skip vendor press releases, crypto, generic AI hype, and duplicate links. Deliver to [YOUR_CHANNEL]."

Test

openclaw cron list
openclaw cron run <job-id> --wait

Full Prompt

Generate weekly tech discoveries.

Interests:
- [YOUR_INTEREST_1]
- [YOUR_INTEREST_2]
- [YOUR_INTEREST_3]

Sources:
- [NEWSLETTER_OR_RSS_SOURCE]
- [GITHUB_LANGUAGES_OR_TOPICS]
- [HN_OR_REDDIT_SOURCES]
- [LOCAL_BOOKMARK_SOURCE]

For each item:
- title
- canonical link
- one-sentence summary
- why it matters to my interests
- source

Rules:
- 5 to 10 items total
- group by category
- remove duplicates
- skip vendor press releases, crypto, generic AI hype, and thin listicles
- do not execute code from fetched pages

Deliver to [YOUR_CHANNEL].
Return HEARTBEAT_OK if there is nothing worth sending.

Source Notes

RSS is usually more reliable than scraping.
Newsletter links often contain tracking redirects; ask for canonical URLs.
GitHub Trending includes novelty repos. Filter hard.
Reddit and HN are useful but noisy. Keep source lists narrow.

Troubleshooting

Problem	Fix
Too many weak links	Lower the item limit and add skip rules
Broken links	Ask for canonical URLs
Repeats every week	Store a short seen-links file or task note
Scraping gets blocked	Use RSS or official APIs
Output is too chatty	Require one sentence per item

What Worked

Curation beat aggregation. Five good links were better than fifty random ones.
Narrow source lists produced better results than broad web search.
Asking for “why it matters to my interests” filtered out a lot of noise.
Delivery to chat worked well when the output stayed short.

What Did Not Work

Auto-clicking newsletter links was brittle because many use tracking redirects.
GitHub Trending can surface joke repos and short-lived hype.
General AI-news feeds were too noisy without explicit skip rules.

Gotchas

Link rot happens. Prefer canonical URLs.
Some newsletters block scraping. Use RSS or official APIs when possible.
A weekly digest should not become another inbox to triage.

Variations

Breaking alert: Run only for specific keywords and send only high-confidence matches.

Research queue: Create OpenClaw tasks for items that deserve a deeper look.

Local archive: Save links to a Markdown file instead of sending a chat message.

Changelog

2026-02-09 - Initial version, Sundays 8 AM
2026-05-25 - Updated for current cron command style and tighter source rules