Fix upstream watch grouping and refresh AI skills

[KSemenenko]

Summary\n- paginate upstream issue discovery and group related watch updates into one open issue per skill set\n- repair dead Agent Framework workflow watch URLs and refresh the upstream baseline state\n- update the Agent Framework, Microsoft.Extensions.AI, and MCP skills for the latest official docs\n\n## Validation\n- python3 -m py_compile scripts/upstream_watch.py scripts/generate_catalog.py\n- python3 scripts/upstream_watch.py --validate-config\n- python3 scripts/upstream_watch.py --sync-state-only\n- python3 scripts/upstream_watch.py --dry-run\n- python3 scripts/generate_catalog.py --validate-only

[Copilot]

Pull request overview

Updates upstream-watch automation to deduplicate related upstream changes into a single open issue per library/skill group (with pagination), and refreshes .NET AI-related skills and watch configuration to align with current official documentation sources.

Changes:

• Group upstream-watch issue creation/update by skill group (issue_key) and paginate open-issue discovery; normalize/repair existing open upstream-watch issues.
• Refresh Agent Framework watch URLs to Microsoft Learn and update upstream baseline state snapshots.
• Update dotnet-microsoft-agent-framework, dotnet-microsoft-extensions-ai, and dotnet-mcp skill docs (including version bumps) for the latest official guidance.

Reviewed changes

Copilot reviewed 8 out of 9 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
skills/dotnet-microsoft-extensions-ai/SKILL.md	Updates skill guidance and bumps version to reflect new .NET AI surface areas (image generation, local models, MCP quickstarts).
skills/dotnet-microsoft-agent-framework/SKILL.md	Updates skill workflow/architecture notes (workflows-as-agent, declarative workflows) and bumps version.
skills/dotnet-mcp/SKILL.md	Updates MCP skill guidance and bumps patch version (quickstarts/registry publishing positioning).
scripts/upstream_watch.py	Implements grouped issue keys/names, payload markers, pagination, reconciliation/closure of duplicates, and grouped upsert behavior.
README.md	Documents that upstream-watch now rolls multiple watches into one grouped upstream issue per library/skill set.
CONTRIBUTING.md	Clarifies grouped issue deduplication behavior for upstream-watch config authors.
AGENTS.md	Adds repo-wide upstream-watch rules: grouping, updating existing issues, and paginating issue discovery.
.github/upstream-watch.ai.json	Repairs Agent Framework watch sources to current Microsoft Learn URLs.
.github/upstream-watch-state.json	Refreshes machine-maintained upstream baseline values (ETags/titles/last-modified/releases).

[Copilot]

parse_open_issue requires both the issue-key marker and a decodable payload before it will treat an issue as a grouped upstream-watch issue. Since the payload already includes issue_key, this means an issue can be silently ignored (and duplicates re-created) if the issue-key marker is missing/edited while the payload is still valid. Consider trusting payload['issue_key'] when present and (optionally) validating it against the marker when both exist.

Suggested change

	if issue_key_match and payload and isinstance(payload.get("watches"), dict):
	issue_key = issue_key_match.group("issue_key")
	if payload and isinstance(payload.get("watches"), dict):
	payload_issue_key = payload.get("issue_key")
	marker_issue_key = issue_key_match.group("issue_key") if issue_key_match else None
	issue_key: str | None = None
	if isinstance(payload_issue_key, str) and payload_issue_key:
	# Prefer the issue key from the payload, but validate against the marker when present.
	if marker_issue_key and marker_issue_key != payload_issue_key:
	# Inconsistent metadata; fall back to legacy parsing for safety.
	return parse_legacy_issue(
	body=body,
	watch_index=watch_index,
	state_watches=state_watches,
	)
	issue_key = payload_issue_key
	elif marker_issue_key:
	# No usable issue_key in the payload; fall back to the marker if available.
	issue_key = marker_issue_key
	if not issue_key:
	# Cannot determine a reliable issue key; treat as legacy.
	return parse_legacy_issue(body=body, watch_index=watch_index, state_watches=state_watches)

[Copilot]

decode_issue_payload calls base64.urlsafe_b64decode directly on the marker contents. If the payload ever loses base64 padding (for example via manual edits/copying), urlsafe_b64decode raises Incorrect padding, causing the issue to be treated as legacy/unknown and potentially breaking deduplication. Consider normalizing padding before decoding (or storing/encoding the payload in a way that is resilient to missing padding).

Suggested change

	decoded = base64.urlsafe_b64decode(match.group("payload").encode("ascii"))
	payload_b64 = match.group("payload")
	# Normalize base64 padding to be resilient to missing '=' characters.
	padding = "=" * (-len(payload_b64) % 4)
	decoded = base64.urlsafe_b64decode((payload_b64 + padding).encode("ascii"))

[Copilot]

parse_legacy_issue calls VALUE_MARKER_RE.search(body) twice in the same expression, doing the regex work twice and making it harder to read. Capture the match once and reuse it when building the fallback snapshot.

Suggested change

	snapshot = {"value": VALUE_MARKER_RE.search(body).group("value")} if VALUE_MARKER_RE.search(body) else {}
	value_match = VALUE_MARKER_RE.search(body)
	snapshot = {"value": value_match.group("value")} if value_match else {}