EPS Dashboard

Overview / Motivation

Auto-filed by /daily 2026-07-23 (transcript sweep). Two Step 10d merges failed on first attempt with "Head branch is out of date" / "Base branch was modified" (#1614 twice; #779's PR #1389 retry) before recovering — on the always-concurrent shared main this race is routine, and each failure burns a merge round.

Goal

Make the Step 10d merge recipe update the head branch BEFORE the first merge attempt (rebase the worktree branch on fetched origin/main, or gh pr update-branch), so the out-of-date first-attempt failure class disappears on busy days.

Workflow gap

Bug observed: the Step 10d recipe attempts gh pr merge --rebase directly; under fleet concurrency (915 commits landed on main on 2026-07-23) the head branch is routinely stale by merge time, so first attempts fail "Head branch is out of date" and the recovery path re-runs the gate machinery.
Why it is a workflow gap: the failure is deterministic-under-load and the pre-update is a one-step recipe addition.
Confidence: medium-high
verified-at-filing: grep -n "update-branch\|out of date" .claude/skills/issue/SKILL.md → 0 hits (absence claim, in-target) (2026-07-24 UTC); two independent same-day incidents in the transcript sweep (#1614 session 26b9e679; #779 session a4b443dd).

Proposed change (refine in planning)

Step 10d: before the first merge attempt, fetch + rebase the issue branch on origin/main (or gh pr update-branch --rebase) and push, THEN merge; keep the existing retry as the backstop.

Scope / surfaces

Primary target: .claude/skills/issue/SKILL.md (Step 10d merge recipe)

Constraints / invariants

Must not weaken the gate ordering (tests run against the landing tree); recursion guard applies.

Provenance

fingerprint: d2798ce57de4
workflow_fix_target: .claude/skills/issue/SKILL.md