harness: reset lists open PRs with --limit 200 and stops if any survive (closes #321)

[mmcky]

Closes #321.

The harness reset's three gh pr list calls had no --limit, so they returned gh's default of 30. A normal gate leaves exactly 28 open PRs per repo, which is why this never showed — but any diagnosis work between two gates pushes a lane past 30, and on 2026-09-21 the .ml lane had 31: the oldest (test-translation-sync.ml#192, from the v0.29.1 gate) was never listed, survived the v0.29.2 reset, and made the lane read 29 PRs / 28 stamped until it was found and closed by hand (tally on #318).

Change

• --limit 200 on all three list calls (source repo, dry-run and live; each target repo).
• A post-reset check: after the close loops, every repo is listed again and the script stops with the leftover PR numbers if any is still open. This catches the class, not just this cause — a listing that fails now counts as a failure rather than reading as "nothing to close" (the target loop's 2>/dev/null || echo "" had made it silent), and a close that did not take is caught too. It runs only outside --dry-run.
• README: one line on the check under "The script will", and its message under Troubleshooting.
• CHANGELOG under [Unreleased].

Checks

bash -n clean; shellcheck warning count unchanged (2, both pre-existing); the leftover-formatting line exercised on a fake two-PR list and on the empty case; workflow-templates.test.ts (the one Jest file that reads the script) green. Not run end-to-end: the harness is currently at @v0 after the v0.29.2 smoke with one open PR per repo, so a run would not exercise the >30 path anyway; the next gate is the first real exercise of the check.

🤖 Generated with Claude Code

[Copilot]

Copilot review overview

🟢 Approval recommended

The changes directly address the documented gh pr list pagination issue and add a sensible safety check without altering core harness behavior beyond improved robustness.

Review effort: Lite
Findings: 1

Open (1)

• gh pr list failures exit without repo context after reset · New

What changed in this PR

This PR hardens the end-to-end test harness reset logic so it reliably closes all open PRs (not just the first 30 returned by gh pr list) and fails fast if any PRs remain open after reset, preventing stale PRs from contaminating gate tallies.

Changes:

• Add --limit 200 to all gh pr list calls used during reset to avoid missing older PRs when a repo has >30 open PRs.
• Add a post-reset verification step (non---dry-run) that re-lists each repo and exits with a clear summary if any open PRs remain.
• Document the new behavior and record the fix in the CHANGELOG under [Unreleased].

File	Description
tool-test-action-on-github/​test-action-on-github.sh	Adds --limit 200 to PR listings and introduces a post-reset “no open PRs remain” assertion to prevent survivors from skewing gate results.
tool-test-action-on-github/​README.md	Notes the new “stop if any PR survives reset” behavior and adds troubleshooting guidance for the new failure mode.
CHANGELOG.md	Records the bugfix and explains the historical failure scenario and the new guardrail.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.