daily-news-digest

Instructions

You will receive a list of sources and topics. For each source:

1. Fetch the index page using web.fetch (JS-rendered if needed).
2. Extract articles published in the last 24h whose title/lede

   matches at least one topic keyword (case-insensitive substring).

3. Collect up to max_items total across all sources, ranked by

   recency.

4. For each selected article, produce a 1-2 sentence summary in

   ptBR (per policies/language.kmd, chatside; UI output stays en-US for product surfaces).

Render the digest in Markdown with this shape:

# Daily digest — {{ now }}

## Top stories ({{ count }})

- **[Source]** [Title](url) — 1-2 sentence summary
  - Why it matters: 1 sentence (only when non-obvious)

## Skipped sources

- Source X — reason (timeout / no matches / paywall)

Then deliver per delivery:

• email → use Koder Mail (when available; until then write to

  {{ workspace }}/digests/{{ date }}.md)

• slack → post to the workspace channel mapped via TOOLS-015
• inline → just return the Markdown string

Steps

1. Validate sources ≥ 1 and each entry has url + name.
2. Parallelise fetches with timeout 30s each.
3. Skip sources that fail (don't abort the whole skill).
4. Rank by published_at desc, cap at max_items.
5. Render Markdown.
6. Deliver per delivery.

Examples

*ser:*(scheduled, daily at 7am) *xpected:*Markdown digest delivered to Slack #daily channel.

*ser:*"digest diário de tech e ia, inline" *xpected:*Markdown returned in-chat, no delivery side effect.

Caveats

• Never paraphrase a paywalled article's body — just title + lede.
• Skip articles older than 24h even if they're "trending".
• If 0 articles match, deliver the digest with a "No matches today"

  message rather than skipping delivery (user expects daily cadence).

References

• AIGW-050 (scheduled tasks — this skill is the canonical example)
• TOOLS-015 (slack adapter)
• policies/language.kmd