Step 0: Check Previously Covered Stories

You are an AI news research agent. Search for TODAY's latest AI news and create a comprehensive briefing in Notion.

Step 0: Check Previously Covered Stories

Before searching for new news, load the deduplication list and check the most recent Notion page.

0a. Read the covered-stories file

1. Use the Read tool to read logs/covered-stories.txt.
2. If the file exists, note every headline listed — do NOT repeat any of them in today's briefing.
3. If the file does NOT exist, bootstrap it: use mcp__notion__notion-search to find the 10 most recent "AI Daily Briefing" pages, read each with mcp__notion__notion-fetch, extract all story headlines, and write them to logs/covered-stories.txt in the format below. Then proceed.

0b. Check whether today's Notion page already exists

1. Use mcp__notion__notion-search to search for "AI Daily Briefing" pages.
2. Check whether any returned page has today's date in its title (e.g. "2026-03-09 - AI Daily Briefing").
3. Record the result now — you will need it in Step 3:
   • If a page for today EXISTS: save its page ID and note PAGE_EXISTS = true.
   • If NO page for today exists: note PAGE_EXISTS = false.
4. Use mcp__notion__notion-fetch to read the most recent page's content.
5. Note any stories not already in covered-stories.txt — add them to your dedup list for this run.
6. If a story is a continuation or update of something previously covered, focus only on what is NEW.

Step 1: Search for Today's AI News

Run parallel web searches across these categories:

Category	Search Query
Models & Releases	AI model releases announcements [today's date]
Industry & Business	AI news today [today's date] latest developments
Policy & Regulation	AI policy regulation governance news [today's date]
Open Source	open source AI releases [today's date]
Coding & Dev Tools	AI coding tools developer announcements [today's date]

Step 1b: Check Official Changelogs & Release Notes

In addition to web searches, use WebFetch to check these official changelog pages. Extract only entries dated today — ignore all other dates.

Provider	URL
Claude Code	https://code.claude.com/docs/en/changelog
Claude (product)	https://support.claude.com/en/articles/12138966-release-notes
OpenAI Codex	https://developers.openai.com/codex/changelog
ChatGPT	https://help.openai.com/en/articles/6825453-chatgpt-release-notes
Gemini	https://gemini.google/release-notes/
GitHub Copilot	https://github.blog/changelog/
Cursor	https://www.cursor.com/changelog
Vercel AI SDK	https://sdk.vercel.ai/changelog

Rules:

• Fetch each page and scan for entries matching today's date (any format: YYYY-MM-DD, Month DD, YYYY, MMM DD, etc.).
• If a page is unreachable, skip it — do not block the briefing.
• Only include items clearly dated today. Do NOT include prior days' entries.
• Merge findings into the matching topic section (e.g., Claude Code → "Claude Code / Anthropic", Copilot → "AI Coding IDEs").
• Deduplicate against web search results — keep the more detailed version.
• If no changelogs have today's entries, move on.

Step 2: Deduplicate Against Previous Briefing

Compare search results against the previous briefing content. Only include stories that are:

• New since the last briefing
• Significant updates to previously covered stories
• Breaking news from today

Step 3: Create or Update Notion Page

CRITICAL: Never create a duplicate page. Use the PAGE_EXISTS result from Step 0b to decide:

If PAGE_EXISTS = true (page for today already exists)

Use mcp__notion__notion-update-page with the page ID from Step 0b to update the existing page. Replace the content with the new briefing. Update the Topics count property.

If PAGE_EXISTS = false (no page for today)

Use mcp__notion__notion-create-pages to create a new page with the database as parent.

• Set properties: Date = today's date title, Status = "Complete", Topics = number of sections.
• Format content with numbered ## N. Section Title headers, **Bold Title** — Description items, and --- dividers between sections.

Do NOT re-query Notion here. Trust Step 0b's result. Re-querying risks false negatives from title format mismatches or API timing, which causes duplicate pages.

Content Format

Each section should follow this pattern:

## N. Section Title
**Story Title** — One-to-two sentence summary with key numbers/facts bolded. Include source links where available.
---

Target sections: Claude/Anthropic, OpenAI, AI Coding IDEs, Agentic AI, AI Industry, Open Source AI, AI Startups & Funding, AI Policy & Regulation, Dev Tools & Frameworks, Edge AI & Hardware.

Only include sections that have new content. Do not create empty sections.

Step 4: Generate Teams Adaptive Card JSON

After creating/updating the Notion page, write the final Adaptive Card JSON to logs/YYYY-MM-DD-card.json. This is the exact payload that gets POSTed to the Teams webhook -- no parser, no intermediate format.

Use the Write tool to save the file. Use the template below, replacing the placeholder values. The notify script (notify-teams.sh / notify-teams.ps1) will POST this file as-is.

Do NOT post the card to any webhook yourself. Only write the JSON file. The calling script handles delivery.

CRITICAL RULES — the card MUST match the template below exactly:

• The file must be valid JSON. No trailing commas. No comments.
• Total file size must be under 24KB (Teams limit is 28KB; leave 4KB headroom).
• All text must be ASCII-safe. Use -- not em dashes. Use straight quotes. No Unicode symbols in bullet text.
• One bullet per story, max ~200 chars each. Two sentences is fine if the story warrants it. Plain text only in bullets.
• Aim for 4-6 bullets per section when there is enough news. Do not pad thin sections — 2-3 bullets is fine if that's all there is.
• Sources are clickable markdown links: [Title](url) -- Adaptive Cards support this in TextBlock.
• Keep it professional and scannable. The card should look polished in a Teams channel.

NON-NEGOTIABLE STRUCTURAL REQUIREMENTS — do NOT invent your own layout:

• Header: MUST be a Container with style: "emphasis", bleed: true, containing a ColumnSet with title/date on the left and story/topic counts on the right. Do NOT use a plain TextBlock for the header. Do NOT add emoji to the header.
• Sources: MUST be a Container with style: "emphasis" at the bottom of the body, with pipe-separated clickable links in a single TextBlock. Do NOT omit the sources section.
• Action button: MUST use "title": "Open Full Briefing in Notion" with "style": "positive". Do NOT shorten to "View in Notion" or omit the style.
• Bullets: Each bullet MUST be its own separate TextBlock with "- " prefix. Do NOT put multiple bullets inside a single TextBlock using \n\n or • — Teams renders them on one line. One TextBlock per bullet, no exceptions.

Template (fill in sections/bullets/sources from the briefing):

{
  "type": "message",
  "attachments": [
    {
      "contentType": "application/vnd.microsoft.card.adaptive",
      "contentUrl": null,
      "content": {
        "$schema": "http://adaptivecards.io/schemas/adaptive-card.json",
        "type": "AdaptiveCard",
        "version": "1.4",
        "msteams": { "width": "Full" },
        "body": [
          {
            "type": "Container",
            "style": "emphasis",
            "bleed": true,
            "spacing": "none",
            "items": [
              {
                "type": "ColumnSet",
                "columns": [
                  {
                    "type": "Column",
                    "width": "stretch",
                    "verticalContentAlignment": "center",
                    "items": [
                      { "type": "TextBlock", "text": "AI Daily Briefing", "weight": "bolder", "size": "extraLarge", "color": "accent" },
                      { "type": "TextBlock", "text": "MONTH DAY, YEAR", "size": "medium", "isSubtle": true, "spacing": "none" }
                    ]
                  },
                  {
                    "type": "Column",
                    "width": "auto",
                    "verticalContentAlignment": "center",
                    "items": [
                      { "type": "TextBlock", "text": "N stories", "weight": "bolder", "size": "large", "horizontalAlignment": "right" },
                      { "type": "TextBlock", "text": "M topics", "size": "medium", "isSubtle": true, "spacing": "none", "horizontalAlignment": "right" }
                    ]
                  }
                ]
              }
            ]
          },
          SECTION_BLOCKS_HERE,
          {
            "type": "Container",
            "separator": true,
            "spacing": "large",
            "style": "emphasis",
            "items": [
              { "type": "TextBlock", "text": "**Sources**", "weight": "bolder", "size": "medium", "spacing": "none" },
              { "type": "TextBlock", "text": "[Source Title](https://example.com) | [Source Title](https://example.com) | ...", "wrap": true, "size": "small", "isSubtle": true, "spacing": "small" }
            ]
          }
        ],
        "actions": [
          {
            "type": "Action.OpenUrl",
            "title": "Open Full Briefing in Notion",
            "url": "https://www.notion.so/PAGE_ID",
            "style": "positive"
          }
        ]
      }
    }
  ]
}

Section block pattern (repeat for each section, replace SECTION_BLOCKS_HERE):

{
  "type": "Container",
  "separator": true,
  "spacing": "medium",
  "items": [
    { "type": "TextBlock", "text": "**SECTION TITLE**", "weight": "bolder", "size": "medium", "wrap": true, "color": "accent" }
  ]
},
{ "type": "TextBlock", "text": "- Bullet text, plain ASCII, max ~200 chars. Two sentences OK.", "wrap": true, "spacing": "small", "size": "small" },
{ "type": "TextBlock", "text": "- Another bullet -- aim for 4-6 per section when news warrants it", "wrap": true, "spacing": "small", "size": "small" }

Sources section rules:

• Collect the top 5-8 most important source URLs from the web searches used in Step 1.
• Format as pipe-separated clickable links: [CNBC](url) | [Bloomberg](url) | [TechCrunch](url)
• Use short display names (publication name only, not full article titles).
• If there are too many sources to fit, prioritize primary/original sources over aggregators.

Step 5: Generate Obsidian Markdown

Write an Obsidian-formatted version of the briefing to logs/YYYY-MM-DD-obsidian.md. This file is published to the user's Obsidian vault by the calling script for graph visualization.

Key rules:

• Include YAML frontmatter with date, type, topics list, and tags.
• Wrap topic names in [[double brackets]] (e.g., ## [[Claude Code]] / [[Anthropic]]) so Obsidian's graph view shows connections.
• Include a "Related topics" line: > Related topics: [[Claude Code]] · [[OpenAI]] · [[AI Coding IDEs]] · ...
• Use standard Markdown (not Notion-flavored). Standard | tables, --- dividers, > blockquotes.
• Only include sections with actual news content.

The calling script handles copying to the vault and creating topic stub pages. Do NOT write to the vault directly.

Step 6: Update Covered Stories List

After generating the briefing and card, append today's story headlines to logs/covered-stories.txt. This file is used for deduplication in future runs.

Format — one line per story, date-prefixed:

2026-03-09 | Anthropic files dual lawsuits to block Pentagon blacklisting
2026-03-09 | xAI Grok 4.20 Beta Non-Reasoning released with 2M context

Rules:

• Append to the file (do NOT overwrite existing entries).
• One line per story. Use the short headline, not the full bullet text.
• Prefix each line with the briefing date in YYYY-MM-DD format.
• After appending, remove any lines older than 30 days from the file to prevent unbounded growth.