add blog monitor skill

CopilotStore/docs-cpp/skills/cpp-blog-monitor/SKILL.md

@@ -0,0 +1,101 @@
+---
+name: cpp-blog-monitor
+description: "Check the Microsoft C++ Team Blog for posts in a requested date range, save unseen C++ DevBlog summaries as Markdown in the temp directory, and avoid reporting the same post twice. Use when the user says: check the C++ blog, C++ DevBlog monitor, summarize C++ blog posts, or what's new on the C++ Team Blog."
+---
+
+# C++ DevBlog monitor
+
+Check the official Microsoft C++ Team Blog and report only posts that haven't
+been successfully summarized in an earlier run.
+
+## Run the monitor
+
+1. If the user hasn't supplied a date range, ask which date range to check.
+   Accept explicit inclusive ranges such as `1-1-2025 to 1-1-2026` and natural
+   ranges such as `this year`, `this month`, `last month`, or
+   `the last two months`. Provide check_cpp_blog.py with a date range converted to the form: mm-dd-yyyy to mm-dd-yyyy 
+
+2. Run the deterministic discovery helper with the user's answer unchanged:
+
+   ```powershell
+   python "${SKILL_DIR}/scripts/check_cpp_blog.py" check --date-range "<date range>"
+   ```
+
+3. Parse the JSON response. If `unseen` is empty, tell the user there are no new
+   posts and stop.
+
+4. For each object in `unseen`, fetch its `url` and summarize the article from
+   the page content. Don't summarize from the RSS excerpt alone.
+
+5. Report each post with this concise structure:
+
+   ```markdown
+   ## [Post title](canonical URL)
+
+   **Published:** Month D, YYYY | **Author:** Name
+
+   Two or three sentences explaining the announcement and why it matters to a
+   C++ developer.
+
+   **Key points**
+
+   - Three to five concrete, technically meaningful points.
+   ```
+
+6. After successfully summarizing a post, pipe its complete Markdown block to
+   the helper. The helper creates or appends to
+   `%TEMP%\BlogSummaries-M-D-YYYY.md`, using the local run date, and prevents
+   duplicate entries for the same post identity. Pass the resolved inclusive
+   endpoints from the `date_range.from` and `date_range.through` fields in the
+   `check` response:
+
+   ```powershell
+   @'
+   <complete Markdown summary block>
+   '@ | python "${SKILL_DIR}/scripts/check_cpp_blog.py" save-summary --post-id "<identity from check output>" --date-from "<date_range.from>" --date-through "<date_range.through>"
+   ```
+
+7. Parse the JSON response and report the returned `output` path to the user.
+   Only after the Markdown summary is saved, record the post as reported:
+
+   ```powershell
+   python "${SKILL_DIR}/scripts/check_cpp_blog.py" record --post-id "<identity from check output>"
+   ```
+
+8. Include the same summary block in the chat response. If fetching,
+   summarizing, or saving one post fails, don't record that post. Continue with
+   the remaining posts and identify the failed URL in the response.
+
+## State and deduplication
+
+The helper stores state at
+`~/.copilot/cpp-blog-monitor/reported-posts.json` by default. It compares both
+the WordPress post identity and canonical URL, so historical API results and
+current RSS results deduplicate against the same records. Writes use a temporary
+file followed by an atomic replacement.
+
+Summary reports are stored separately in the operating system's temp directory
+as `BlogSummaries-M-D-YYYY.md`. Multiple successful summaries on the same local
+date are appended to that day's file. Each distinct resolved inclusive run
+range appears directly below the report's **Generated** line. Both report and
+ledger writes are atomic.
+
+Explicit range endpoints are inclusive. `this year` and `this month` start at
+the corresponding calendar boundary and end today. `last month` is the previous
+complete calendar month. `the last N months` is a rolling range from the same
+day N months ago through today.
+
+Override the state file for testing or isolated runs with `--ledger <path>`.
+Use `--limit <number>` on `check` only when the user requests a maximum number
+of posts.
+
+## Boundaries
+
+- This skill runs on demand. It doesn't schedule itself or continuously poll.
+- Historical discovery uses the blog's paginated WordPress API, not the
+   10-entry RSS window.
+- Use Windows Task Scheduler, cron, or another automation service for unattended
+  checks.
+- Don't record a post before its summary succeeds.
+- Don't record a post before its Markdown summary is saved.
+- Don't modify or clear the ledger unless the user explicitly asks.