Daily News Digests

Use this skill when the user asks for a recurring local news/research monitor that pulls fresh items from X/Twitter, GitHub, Reddit, Hacker News, RSS, release notes, forums, or other public sources and saves a Markdown digest locally.

The deliverable is a working collector plus verified scheduling, not a suggestion list.

Default outcome

Create:

1. A deterministic script under ~/scripts/ or an explicitly requested project folder.
2. A dated Markdown output directory under the relevant vault area.
3. A latest.md copy for quick viewing.
4. A scheduler:

• macOS local/background: launchd LaunchAgent.
• Hermes/gateway-delivered work: Hermes cron only when delivery or LLM reasoning is needed.

5. A verified test run with real output, logs, and exit status.

Source selection pattern

Prefer primary or near-primary feeds before broad web search:

1. Official GitHub repo:

• releases
• issues
• pull requests
• commits
• GitHub search
• Atom feeds for releases/commits

2. Official docs/blog/release notes RSS or sitemap if available.
3. X/Twitter via xurl when installed and authenticated.
4. Reddit public JSON search for relevant subreddits and global mentions.
5. Hacker News Algolia search.
6. Other forums/community sources only if the topic warrants it.

For X/Twitter, do not read or print ~/.xurl. Check only xurl auth status; if auth is unavailable, make the script skip X cleanly and explain how the user can enable it outside the agent session.

Implementation steps

1. Decide output path based on domain ownership.

• Assistant/system monitoring: ~/Notes/<Topic>/
• Personal: ~/Notes/...
• Work/finance/safety domains: delegate or ask the owning profile if needed.

2. Write a standalone script with:

• Configurable window, e.g. HERMES_NEWS_DAYS.
• Configurable output directory, e.g. HERMES_NEWS_DIR.
• Configurable max items per source.
• Per-source exception handling so one failing source does not kill the digest.
• Deduplication by canonical URL.
• Markdown grouped by source.
• A clear note that collection is not verification.

3. Run syntax validation:

• Python: python3 -m py_compile <script>.

4. Run the script manually and verify:

• Exit code is 0.
• Dated Markdown file exists and is non-empty.
• latest.md exists and is non-empty.
• Output includes real collected item counts.

5. Schedule it.
6. Kickstart/run the scheduled job once and verify scheduler logs.

macOS launchd pattern

Use LaunchAgents for per-user local Markdown collectors:

• File: ~/Library/LaunchAgents/<reverse-domain-label>.plist
• Validate: plutil -lint <plist>
• Load/reload:

UID_NUM="$(id -u)"
launchctl bootout "gui/$UID_NUM/<label>" 2>/dev/null || true
launchctl bootstrap "gui/$UID_NUM" "$HOME/Library/LaunchAgents/<label>.plist"
launchctl enable "gui/$UID_NUM/<label>"
launchctl kickstart -k "gui/$UID_NUM/<label>"

• Verify:

launchctl print "gui/$(id -u)/<label>"
tail -50 ~/Library/Logs/<name>.log
tail -50 ~/Library/Logs/<name>.err.log

Include a robust PATH in the plist EnvironmentVariables so Homebrew/user tools are found:

<key>PATH</key>
<string>/opt/homebrew/bin:/usr/local/bin:/usr/bin:/bin:/usr/sbin:/sbin:$HOME/.local/bin</string>

Hermes cron vs launchd

Use launchd when:

• The task is local-only.
• It writes files locally.
• No LLM reasoning or gateway delivery is required.
• the user asked to “handle it” and “test it.”

Use Hermes cron when:

• The job should summarize, classify, or decide using an LLM each run.
• The output should be delivered to Telegram/Discord/etc.
• The job should chain from another Hermes cron job.

CLI sessions have no live delivery channel. Do not promise that default deliver='origin' cron jobs will message the user in the terminal.

Verification checklist before final response

Report the actual evidence:

• Script path.
• Scheduler plist/job path.
• Output Markdown path.
• Last scheduler exit code or equivalent status.
• Log paths.
• Count of collected items if available.
• Whether stderr is empty.

Avoid saying “it should run.” Say either “it is loaded and tested” or state the blocker.

Pitfalls

• Do not stop at a plan; create and run the collector.
• Do not let X/Twitter auth failure break the whole digest. Skip X with a note.
• Do not print credentials or auth files.
• Do not treat collected social posts as verified facts; mark the digest as collection-only.
• When a terminal command exits non-zero because a final diagnostic command was missing or shell-specific, inspect the useful output first. If service status and logs show success, do not over-correct a working scheduler.