morning-report

══════════════════════════════════════════════════════════ MORNING REPORT SKILL #1 — weather + daily news briefing: ══════════════════════════════════════════════════════════ Recognize a request for a morning briefing — "morning report", "morning briefing", "daily brief", "give me my morning update", "weather and news summary", or similar — as a weather+news briefing YOU own end-to-end. This is a DATA-DEPENDENT chain (see the COMPOUND TURN RULE box): fetch the raw inputs first with read-only shell commands, WAIT for their results, then YOU compose, deliver, and offer the schedule (steps 3–5). Do NOT call assistant_handoff and do NOT start an investigation/gather for this request — that produces a second, unrelated "status report" after the briefing. Never fabricate weather values or headlines, and never emit the compose/deliver step in the same response as the fetches.

OUTPUT MUST BE HUMAN-READABLE — NEVER RAW MARKUP. The fetched feed is raw RSS/XML/HTML; it is INTERMEDIATE data only. NEVER present the raw feed, XML tags, CDATA blocks, or angle-bracket markup to the user, and NEVER let a raw curl dump be the final answer. The user only ever sees the composed plain-text briefing produced in the final step. The news fetch below already strips the feed down to plain-text headline lines so nothing but readable text comes back. STEP LABELING RULES (UX) — the fetches below run quiet, so narrate the 5 steps:

• Before every step's tool calls, emit this exact header format as assistant text in the SAME response as the tool calls, then one short status sentence:

  [n/5]

• Steps 1–2 fire as one parallel batch; label that batch with the combined header ### [1-2/5] Fetch weather + headlines.
• After a step's tool results are in, state its outcome in one line (start it with ✓ on success, ✗ plus what failed otherwise) before the next step's header (e.g. ### [4/5] Deliver to Slack).
• Use each step's own number as n; never renumber mid-run. The composed briefing itself (step 3) and the schedule offer's response_text (step 5) stay exactly as specified below — headers narrate around them, never replace them.

Steps, in order:

1. Fetch today's weather with shell_run. Use quiet=true so the $ curl line and raw stdout stay off the user's screen — they only need the composed briefing in step 3, not the same weather line twice. Use the city the user named; if none is given, use their configured/default city, else omit the location: → shell_run(command="curl -s 'wttr.in/?format=%l:+%c+%t,+feels+%f,+%h+humidity,+wind+%w'", quiet=true)
2. Fetch current headlines with shell_run as PLAIN TEXT — extract just the headline titles from the feed, drop the channel title, and cap the list. Also quiet=true (same reason: intermediate fetch, not the user-facing answer). Do NOT fetch the raw feed without this extraction pipeline: → shell_run(command="curl -s 'https://feeds.bbci.co.uk/news/rss.xml' | grep -oE '<title><!\[CDATA\[[^]]*\]\]>' | sed -E 's/<title><!\[CDATA\[//; s/\]\]>//' | sed '1d' | head -n 8", quiet=true)
3. After BOTH tool results are in, compose a clean, human-readable briefing from the ACTUAL fetched data. Required format — Markdown/plain text only, no HTML/XML, no links, no angle brackets: Good morning! Here is your briefing. Weather — : <one-line conditions from step 1> Top headlines:
   • <headline 1, one short sentence>
   • <headline 2>
   • ... (3–5 bullets total)
4. ALWAYS DELIVER TO SLACK — as the FINAL action of this skill you MUST send the WHOLE composed briefing (both the weather/temperature line AND the news headlines, exactly as formatted in step 3) to Slack via slack_send_message, even when the user did NOT explicitly ask to send it anywhere. This is a DATA-DEPENDENT step: emit it on the response AFTER the two fetches, with the full composed plain-text briefing in message — never raw feed markup, never a partial report, never a "preparing…" placeholder. The Slack webhook is bound to a single preconfigured channel, so do NOT ask which channel to use. If the user names another platform (e.g. "post it to telegram"), ALSO deliver there with telegram_send_message; Slack stays the default sink. Skip a platform only if it is not connected. If neither delivery tool is available, defer to the "Delivery tool unavailable" rule above instead of fabricating a command. Examples:

• "give me my morning report" → shell_run(command="curl -s 'wttr.in/Amsterdam?format=%l:+%c+%t,+feels+%f,+%h+humidity,+wind+%w'", quiet=true) + shell_run(command="curl -s 'https://feeds.bbci.co.uk/news/rss.xml' | grep -oE '<title><!\[CDATA\[[^]]*\]\]>' | sed -E 's/<title><!\[CDATA\[//; s/\]\]>//' | sed '1d' | head -n 8", quiet=true) [both this turn — independent fetches] → (observe both results) → slack_send_message(message="<the FULL composed plain-text weather + headlines briefing>") [next turn — mandatory Slack delivery]
• "morning briefing for Berlin and post it to telegram" → shell_run(command="curl -s 'wttr.in/Berlin?format=%l:+%c+%t,+feels+%f,+%h+humidity,+wind+%w'", quiet=true) + shell_run(command="curl -s 'https://feeds.bbci.co.uk/news/rss.xml' | grep -oE '<title><!\[CDATA\[[^]]*\]\]>' | sed -E 's/<title><!\[CDATA\[//; s/\]\]>//' | sed '1d' | head -n 8", quiet=true) [both this turn] → (observe both results) → slack_send_message(message="<the FULL composed briefing>") + telegram_send_message(message="<the FULL composed briefing>") [next turn — Slack always + Telegram because the user named it]

5. AFTER the briefing exists (steps 1–3 done; step 4 delivery attempted), ALWAYS offer to make mornings recurring. NEVER call propose_scheduled_delivery as the first or only tool of this skill — the tool will REJECT that and the user would only see a schedule offer with no weather/news. Steps 1–3 must have run first. Call propose_scheduled_delivery with briefing_text set to the FULL composed briefing from step 3 (required). The tool returns response_text = briefing + closer; show that to the user (or at least end with the closer). Do NOT call /cron yet; the user's bare "yes" expands to the tool's slash_preview with no LLM round-trip. Defaults when they accept without overrides: weekdays 08:00 in their timezone if known else UTC, provider matching where you just delivered (Slack webhook by default — omit chat_id). Kind is daily_summary. Example after Slack webhook delivery: → propose_scheduled_delivery(kind="daily_summary", cron="0 8 * * 1-5", timezone="UTC", provider="slack", briefing_text="<FULL composed weather + headlines briefing>") → show the tool's response_text (briefing + Want me to: …) Pass chat_id only when you have a concrete Telegram/Discord/Rocket.Chat destination (or a Slack #name/C… you already reported). Never invent one. Skip the offer only when they already asked for a one-off and explicitly declined recurrence earlier in this conversation. A briefing that runs once is a demo; a morning that arrives every weekday is the product.