News Brief Automation

Build, refine, or maintain recurring news-digest workflows that periodically collect items from sources such as Hacker News, GitHub Trending, Hugging Face, RSS feeds, blogs, and specific news sites; deduplicate against the newest prior report; save a Markdown brief in the workspace; and send the same summary to the user. Use when asked to set up scheduled news checks, daily/weekly digests, trend briefings, cron prompt files, or reusable report automation for “check news and send me a summary” tasks.

Turn ad-hoc “go check the news and summarize it” work into a repeatable automation made of:

• one prompt file under automation/cron/
• one report folder under reports/
• one stable output format
• one explicit dedupe rule against the latest saved report

Workflow

1. Identify the recurring job:
   • source set
   • cadence
   • selection criteria
   • output language
   • save location
   • delivery target
2. Prefer one prompt file per job under automation/cron/<job-name>.md.
3. Create a matching report directory under reports/<topic>/.
4. Instruct the agent to compare candidates against the newest saved report before drafting the final output.
5. Require “save first, send second” so future runs always have a concrete baseline for dedupe.
6. Make the format stable and easy to skim.
7. If a run has little novelty, emit a short incremental edition instead of padding the report.

What A Good Cron Prompt Must Contain

Each recurring prompt should explicitly define:

• the goal
• the schedule context
• the priority sources to inspect
• what counts as worth including
• the required output structure
• dedupe rules against the newest prior report
• save path and filename convention
• send-after-save behavior
• fallback behavior if a source is unavailable

Use direct instructions. Avoid fuzzy wording like “make it good” or “summarize nicely”.

Good Defaults

Source handling

Prefer primary index pages first, then open individual links only when context is missing.

Common defaults:

• Hacker News: homepage + newest
• GitHub: Trending
• Hugging Face: hot/trending models, datasets, spaces
• RSS/blogs: feed page first, article page second
• Generic news sites: homepage/section page first, article page second

Selection logic

Prioritize items that are:

• newly surfaced
• rapidly rising
• widely discussed
• practically useful
• unusually opinionated or technically deep

For technical audiences, bias toward:

• AI
• developer tools
• open source
• infrastructure
• security
• data engineering
• engineering practice

Deduplication

Always inspect the newest report in the target report folder.

Keep only:

• new items
• items with clear rank/momentum change
• items with meaningful new developments
• previously seen items that deserve a shorter refreshed mention

Drop or compress:

• unchanged repeats
• low-signal filler
• items included yesterday/last run with no meaningful change

If most candidates repeat prior coverage, explicitly publish an incremental short edition.

Output structure

Prefer Markdown with:

• dated title
• short “what matters now” section
• numbered list of items
• link for each item
• one-line reason to care
• optional discussion angle / watchpoint

Keep tone concise and opinionated enough to be useful.

File And Naming Conventions

Use this layout unless the user asks otherwise:

text

automation/cron/<job-name>.md
reports/<topic>/YYYY-MM-DD.md
reports/<topic>/YYYY-MM-DD-HH00.md

Use date-only filenames for once-per-day reports. Use hour-stamped filenames for multiple runs per day.

Common Patterns

Daily digest

Use when the user wants one higher-signal summary per day.

Recommended traits:

• 5-10 items
• more synthesis
• a “today’s observation” section
• date-only filenames

Intraday brief

Use when the user wants multiple updates per day.

Recommended traits:

• 3-8 items
• stronger dedupe
• incremental mode when nothing major changed
• hour-stamped filenames

Mixed-source trend watch

Use when combining 2-4 complementary sources like GitHub + Hugging Face + HN.

Recommended traits:

• sectioned output per source
• final cross-source observations
• avoid repeating the same project across sections unless the angle differs

Editing Existing Automations

When converting an existing one-off prompt into a reusable workflow:

1. Preserve the parts that already work:
   • source order
   • output shape
   • report path
2. Remove accidental details that do not generalize.
3. Keep the reusable operating logic:
   • fetch
   • select
   • dedupe
   • save
   • send
4. Move concrete examples and templates into references/ so SKILL.md stays lean.

Quality Bar Before Packaging Or Publishing

Before packaging or publishing:

• keep SKILL.md concise and trigger-focused
• ensure the description clearly says when the skill should activate
• provide at least one reusable prompt template in references/
• avoid extra docs like README or changelog files inside the skill
• package the exact skill folder
• reject vague prompts that do not specify dedupe or save paths

References

• Read references/examples.md for concrete example automations.
• Read references/templates.md for reusable cron prompt templates.