[elbruno]

Replaces the local Copilot CLI schedule with a repo-hosted monitor, so tracking no longer depends on a developer machine being online.

What it watches

ggml-org/llama.cpp#27742 - model: add Qwen3.8-Flash-Next (qwen4exp). That PR is currently the blocker for the GGUF path (ONNX is blocked separately and more fundamentally - see #53).

Behaviour

Upstream state	Action
PR open (today)	Silent success - run summary only, no issue noise
PR merged	Comments once on #53 with merge details, llama.cpp release, and LLamaSharp status
PR closed unmerged	Comments once on #53 flagging that a new tracking target is needed

Design notes

Follows the conventions established by the (now retired) monitor-gemma4-blocker.yml:

• Injection-hardened - all config passed via env: and read through process.env, never interpolated into the script body. This was a lesson learned in that workflow's history (commit \d979fc2).
• Silent by default - a daily job that comments daily is a job people mute. It writes only to the run summary until something actually changes.
• Marker-based idempotency - terminal announcements carry an HTML comment marker and the workflow scans existing comments before posting, so a merge is announced exactly once rather than every morning thereafter.
• Terminal-state pattern - the header documents retiring this to a manual-only stub once resolved, matching how the Gemma 4 monitor was wound down.

One deliberate caution

The workflow does not claim C# usability when the PR merges. LLamaSharp pins a specific llama.cpp commit and ships its own native binaries, so an upstream merge is necessary but not sufficient. The LLamaSharp release date is reported as an explicitly-labelled heuristic with a checklist item to verify the pinned commit - rather than asserting support that may not exist yet.

Validation

• YAML parsed with yaml.safe_load - OK
• Embedded github-script JS extracted and syntax-checked with node --check - OK
• No repo secrets required; uses the default GITHUB_TOKEN with issues: write

Refs #53