name	morning
description	Daily 06:30 BKK compile — ingest yesterday's raw, research flagged links, write daily summary, conditionally weekly/monthly, push Telegram digest, open PR.

AgenticWiki — Morning Skill

Single skill that does the entire morning compile. Invoked by the Cowork Routine morning-routine. Runs once per day at 06:30 Asia/Bangkok.

Hard rules:

Read CLAUDE.md first — it is the schema for how the wiki must be maintained.
Never auto-merge any PR.
Never edit raw/, wiki/log.md past entries, or files in followups/done/.
If a step fails, log the failure in the PR body and continue with the rest.

Step 1 — Date scope (Asia/Bangkok)

Compute now:

TODAY = today's date YYYY-MM-DD
YESTERDAY = TODAY - 1 day
IS_SUNDAY = today's weekday is Sunday
IS_MONTH_END = (today + 1 day) is in a different month — i.e. today is the last day of its month

Cite these values in the PR body so the human reviewer sees what you assumed.

Step 2 — Ingest yesterday's raw

Read everything created or modified yesterday:

raw/sessions/{YESTERDAY}/ — Telegram bot dumps (text, voice, link)
raw/voice/{YESTERDAY}.md — LINE voice log (if exists)

Special file: raw/sessions/{YESTERDAY}/daily-log-{YESTERDAY}.md — if the operator typed /eod to close out the day, this marker file exists. Treat its ## Note section (if present) as the operator's own bottom-line for the day, and incorporate it as the seed for the daily summary in Step 4.

Session grouping: every dump file has frontmatter fields session (e.g. s-12345) and seq (1, 2, 3, …). Files sharing the same session were sent within 5 minutes of each other and almost always reference each other linguistically (e.g. one starts with "อ่อแล้วก็…" or just continues the previous thought). When extracting entities or quoting, read the entire session in seq order before deciding what's a coherent thought vs. a separate one. Treat a session as one continuous brain-dump split across multiple Telegram messages.

For each entity mentioned (person, company, product, concept, event):

If a wiki/ page exists → propose update (add facts, citations, [[wikilinks]])
If not → create a stub with frontmatter + 1-paragraph summary + sources
Connect new content with [[wikilinks]] to existing pages

Append one block to wiki/log.md (append-only, never edit prior entries):

## {TODAY} 06:30 — Morning routine
- Read: raw/sessions/{YESTERDAY}/*, raw/voice/{YESTERDAY}.md
- Created: wiki/...
- Updated: wiki/...
- Followups extracted: N

For lines tagged "อยากดูต่อ" / "น่าสนใจ" / "ต้อง research" / forwarded articles / URLs → create one file per item under followups/pending/{slug}.md with the question + source citation.

Step 3 — Research flagged items

For every entry in:

raw/sessions/{YESTERDAY}/*-link.md (URLs)
followups/pending/*.md (open questions)

do this:

Web fetch the URL (or web search if it's a question, not a URL)
Synthesize a clear, easy-to-read article in Thai — 3–6 paragraphs, the tone of a reflective tech blogger explaining to a smart friend:
- Define jargon when first introduced
- No hype, no marketing voice
- Always cite the source URL
Save the article:
- If it's a new entity → wiki/{type}/{slug}.md (concepts, people, companies, events)
- If the topic already has a wiki page → append a new section there with today's date as a subheading
Move the followup file followups/pending/{slug}.md → followups/done/{slug}.md and add researched: {TODAY} to its frontmatter

Step 4 — Daily summary

Write a reflective first-person Thai recap to outputs/daily/{YESTERDAY}.md. Address the operator directly ("เมื่อวาน คุณ..."). Tone: warm, honest, unhurried — like a journal entry written by someone who actually knows them.

Length: 4–8 paragraphs. Use these exact subheadings:

บันทึกของเมื่อวาน — what you actually said in voice/text, paraphrased
Wiki ที่งอกเพิ่ม — N pages created, M updated, top 3 most-connected
Research ที่ส่งให้แล้ว — links you flagged → tl;dr of each, with inline link to the wiki page
TODO health check — read todos/open/*.md. Report:
- counts by tag (HMN N · ENB M · PERS K)
- overdue (due_at past today): list with original date
- coming up next 24h: list with due_at
- completed yesterday (from todos/done/*.md with completed_at matching YESTERDAY): count by tag
- Stale flag: any open TODO with created more than 7 days ago → surface with note "consider /done or /cancel_todo"
Patterns ที่สังเกตเห็น — recurring themes, what you obsessed over
Seed สำหรับวันนี้ — 1–2 questions worth following up today

Step 5 — Conditional: weekly summary

If IS_SUNDAY, generate outputs/weekly/{YYYY-WW}.md covering the past 7 days (Mon–Sun ending yesterday). Sections per CLAUDE.md "Weekly report format":

What happened in your wiki this week
Industry moves you noted
What you were curious about
Contradictions surfaced (if any)
Narrative — 2–3 paragraphs on the shape of your thinking this week

Step 6 — Conditional: monthly summary

If IS_MONTH_END, generate outputs/monthly/{YYYY-MM}.md for the current month. Reflective + analytical:

Themes that defined the month
"What you changed your mind about" — explicit list with old → new
Drift — topics you used to pay attention to but stopped
Top 5 wiki pages by inbound links
One paragraph: where you're heading

Step 7 — PR + Telegram digest

7a. Open Pull Request

Commit all changes onto a new branch morning/{TODAY} and open a PR titled:

morning: {TODAY} — N pages, M research[, weekly][, monthly]

PR body:

Punch list of all created/updated/moved files (grouped by section)
The date assumptions from Step 1
Any failures or skipped sub-steps with reason

Do not auto-merge. Operator reviews on mobile and merges manually.

7b. Telegram digest push

Send a SHORT message to the operator's Telegram (≤ 4000 chars). Format:

📅 *Morning brief — {TODAY}*

📝 เมื่อวาน:
{2–3 sentence paraphrase of yesterday's bottom-line — what stood out}

📚 Wiki: {N} new, {M} updated
🔍 Research: {K} items delivered
{if IS_SUNDAY:    📊 Weekly summary ready}
{if IS_MONTH_END: 🗓️ Monthly summary ready}

PR: {pr_url}
อ่านเต็ม: outputs/daily/{YESTERDAY}.md

Send via Telegram Bot API. Use whichever HTTP/code-execution tool you have. Equivalent of:

curl -X POST "https://api.telegram.org/bot${TELEGRAM_BOT_TOKEN}/sendMessage" \
  --data-urlencode "chat_id=${TELEGRAM_OWNER_CHAT_ID}" \
  --data-urlencode "parse_mode=Markdown" \
  --data-urlencode "text=$MESSAGE"

Verify the response is {"ok": true, ...}. If false, log to PR body but don't fail the run.

Done

End the routine.

name	morning
description	Daily 06:30 BKK compile — ingest yesterday's raw, research flagged links, write daily summary, conditionally weekly/monthly, push Telegram digest, open PR.