watch-pr

# Watching a PR and handling the Copilot review ## CI checks Wait for checks in the background (never foreground-sleep): ```bash # run with run_in_background: true until [ "$(gh pr checks <N> -R shopsys/helm 2>/dev/null | grep -c pending)" = "0" ]; do sleep 10; done gh pr checks <N> -R shopsys/helm | awk -F'\t' '{print $1": "$2}' ``` A stacked PR (base ≠ main) gets NO checks until it is retargeted to main — that is expected, not a failure. ## Fetching the Copilot review Copilot feedback arrives in TWO places; always check both: ```bash # 1. regular review comments (threads) - top-level AND replies gh api repos/shopsys/helm/pulls/<N>/comments \ --jq '.[] | "id: \(.id) | reply_to: \(.in_reply_to_id) | \(.path):\(.line // .original_line)\n\(.body)"' # 2. SUPPRESSED comments - only inside the review body, no threads exist for them gh api repos/shopsys/helm/pulls/<N>/reviews \ --jq '.[] | select(.user.login | test("copilot"; "i")) | .body' | grep -A30 "Suppressed" ``` ## Triage protocol For every finding, decide explicitly and act: - **Valid → fix**: implement, add a test covering it, run the full verification suite, push, then reply in the thread referencing the commit short-SHA. - **Invalid or intentional → justify**: reply in the thread with the concrete reasoning (link the upstream issue/doc that proves the point). If the behavior is intentional, also document it in the code/values comment or the deviations doc so the next reviewer doesn't re-raise it. - **Suppressed comments**: no thread exists — triage them in a single PR conversation comment (`gh pr comment`) listing each finding and its resolution. Replying to a thread: ```bash gh api repos/shopsys/helm/pulls/<N>/comments/<comment-id>/replies -f body="..." ``` **NEVER resolve review threads yourself** — leave them open so Copilot can respond; the maintainer resolves them (or asks for it explicitly). Copilot replies inside threads have `in_reply_to_id` set — check for them on follow-up passes. Copilot re-reviews after every push; expect a new review (possibly with new suppressed comments) after each fix round and repeat the triage. ## Long-running watch across PRs For a persistent watch over all open PRs (new Copilot comments AND thread replies, suppressed-comment reviews, CI failures, merges), use the Monitor tool with a poll loop over `gh api` (state file with seen IDs; poll ≥ 90 s). Caveat: seed the state file with existing IDs at start, and beware the init race — a review submitted during initialization gets seeded as already-seen, so after (re)starting the monitor do one manual pass over the PRs.