diff --git a/.claude/agents/.AGENTS_PROVENANCE b/.claude/agents/.AGENTS_PROVENANCE
index 545411de..5ed9fd3b 100644
--- a/.claude/agents/.AGENTS_PROVENANCE
+++ b/.claude/agents/.AGENTS_PROVENANCE
@@ -1,6 +1,6 @@
 # VENDORED from harmon-devkit — DO NOT EDIT the managed agents here.
 # source: https://github.com/evanharmon1/harmon-devkit.git
-# ref: v0.29.0 (e6596fb0fa1890a187e26ed9aae170517960d861)
+# ref: v0.30.0 (92451b5b419779ef7823ea8acaaad4826ffebd22)
 # path: ai/agents
 # names: *
 # managed: implementer
diff --git a/.claude/skills/.SKILLS_PROVENANCE b/.claude/skills/.SKILLS_PROVENANCE
index bf842870..8eefe0de 100644
--- a/.claude/skills/.SKILLS_PROVENANCE
+++ b/.claude/skills/.SKILLS_PROVENANCE
@@ -1,6 +1,6 @@
 # VENDORED from harmon-devkit — DO NOT EDIT the managed skills here.
 # source: https://github.com/evanharmon1/harmon-devkit.git
-# ref: v0.29.0 (e6596fb0fa1890a187e26ed9aae170517960d861)
+# ref: v0.30.0 (92451b5b419779ef7823ea8acaaad4826ffebd22)
 # path: ai/skills
 # categories: universal, repo
 # agents-dest: .claude/agents
diff --git a/.claude/skills/shepherd/SKILL.md b/.claude/skills/shepherd/SKILL.md
index 71a4c3b2..17b4862b 100644
--- a/.claude/skills/shepherd/SKILL.md
+++ b/.claude/skills/shepherd/SKILL.md
@@ -700,8 +700,12 @@ you rather than the bot):
   reply to, so no act on GitHub can ever record that you answered it and
   `check` returns exit 10 for that head forever — the deadlock the inline
   adjudication path was built to end, reappearing on the two surfaces it
-  cannot see. Answer the finding on the PR as usual (fix it, decline it with
-  reasoning, or file it), then record the disposition:
+  cannot see. Answer the finding on the PR as usual — and note that only two
+  of the three answers end here. **Fixing** it means a push, which moves the
+  head and starts a fresh cycle that reviews the fix on its own merits;
+  `settle` neither applies nor accepts that disposition. For the two answers
+  that leave the code alone — declining with reasoning, or filing it as
+  follow-up work — record the disposition:
 
   ```bash
   "$helper" settle --state "$state" --actor-id 199175422 \
@@ -1262,17 +1266,21 @@ loops indefinitely:
    When current-head Codex cloud review is enabled, **Codex Automatic reviews
    must be disabled in the external integration before the first promotion**.
    Otherwise `gh pr ready` can start a new asynchronous review after the gate
-   that supposedly completed automated work. Three knobs, all of them:
+   that supposedly completed automated work. Three knobs carry it:
    personal **Auto review** off, the repository's **Auto code review**
-   preference on **Follow personal**, and the repository's review **Trigger**
-   on Follow personal — an "On every push" trigger is dormant while Auto
-   review is off and arms the moment that toggle changes. GitHub does not
-   expose a reliable
-   repository API for this setting, so treat it honestly as a human-configured
-   prerequisite: use the repository setup record or maintainer confirmation,
-   never claim it was mechanically verified — a passing `readiness-gate.sh`
-   deliberately does not check it, and says so. If its state is unknown, stop
-   blocked and ask rather than promote.
+   preference on **Follow personal**, and the repository's
+   review **Trigger** on Follow personal — an "On every push" trigger is
+   dormant while Auto review is off and arms the moment that toggle changes.
+
+   **This is settled configuration, not a promotion-time check.** The
+   consuming repository's `AGENTS.md` carries the maintainer's confirmation
+   and its setup checklist carries the how-to; nothing in this stage gates on
+   it, and `readiness-gate.sh` says nothing about it either.
+   The one thing worth raising is an anomaly you happen to observe: if a
+   Codex cloud review fires **unsolicited** — after a push or a promotion
+   that no `@codex review` comment triggered — tell the maintainer, because
+   that is the signature of the knobs drifting back on. Report it and carry
+   on; it blocks nothing and there is no state to poll.
 
    Report the ready state honestly rather than over-claiming:
    `BLOCKED` or `REVIEW_REQUIRED` mean "ready for review and awaiting the
diff --git a/.claude/skills/shepherd/assets/readiness-gate.sh b/.claude/skills/shepherd/assets/readiness-gate.sh
index ba6925eb..2bacc90c 100755
--- a/.claude/skills/shepherd/assets/readiness-gate.sh
+++ b/.claude/skills/shepherd/assets/readiness-gate.sh
@@ -55,13 +55,11 @@
 #   checks-indeterminate, merge-state-unknown, fetch-failed,
 #   malformed-data, codex-indeterminate, usage              (indeterminate)
 #
-# Three readiness conditions are deliberately NOT verified here, because no
+# Two readiness conditions are deliberately NOT verified here, because no
 # API answers them — the caller must hold them as prose prerequisites:
 #   - required automation that reacts only to `pull_request.ready_for_review`
 #     (a configuration blocker: promotion would notify humans before its
 #     result exists);
-#   - the Codex Automatic-review knobs being off/Follow-personal (a
-#     human-configured prerequisite; GitHub exposes no reliable API for it);
 #   - a required context that NEVER REGISTERED on this head. The checks
 #     condition judges every check GitHub reports for the commit; a required
 #     workflow that failed to trigger appears in no list, and the only state
@@ -411,9 +409,19 @@ evaluate_checks() {
     # suites asynchronously, so a read moments after a push reports nothing
     # having run rather than nothing to run. A repo with genuinely no CI
     # needs a human to say so — this gate cannot tell the two apart.
+    # --slurpfile over process substitution, never --argjson: a much-rerun
+    # head's check-runs payload exceeds the per-argument limit as argv and jq
+    # dies "Argument list too long", which this gate could only report as
+    # `malformed-data` — indeterminate for a purely mechanical reason, on
+    # exactly the heads it matters most for (observed on harmon-init#821's
+    # gate after three infra reruns, 2026-08-12, where it also masked a real
+    # merge-state-behind condition). printf is a shell builtin, so no exec
+    # carries the payload; the fd does. $runs/$statuses are bound below so the
+    # classification program itself is unchanged.
     checks_summary="$(jq -cn \
-        --argjson runs "$check_runs" \
-        --argjson statuses "$statuses" '
+        --slurpfile runs_sf <(printf '%s' "$check_runs") \
+        --slurpfile statuses_sf <(printf '%s' "$statuses") '
+          $runs_sf[0] as $runs | $statuses_sf[0] as $statuses |
           def run_state:
             if .status != "completed" then "pending"
             elif (.conclusion == "success" or .conclusion == "neutral"
@@ -432,10 +440,21 @@ evaluate_checks() {
         indeterminate malformed-data "check states could not be classified"
     [ "$(jq -r '.total' <<<"$checks_summary")" -gt 0 ] ||
         indeterminate checks-indeterminate "GitHub reports no checks for this head — populated asynchronously, so re-poll; if the repo truly has no CI, that is a human call, not a pass"
-    failing_checks="$(jq -r '.failing | join(", ")' <<<"$checks_summary")"
+    # Bound the name lists the detail carries: with thousands of failing
+    # checks the joined names are themselves a multi-megabyte string, and
+    # emit's `jq --arg detail` puts that back into a single argv entry — the
+    # same ARG_MAX death the --slurpfile change above just removed, one step
+    # downstream. Twenty names diagnose as well as twenty thousand.
+    failing_checks="$(jq -r '
+        .failing | if length > 20
+        then (.[0:20] | join(", ")) + " … and \(length - 20) more"
+        else join(", ") end' <<<"$checks_summary")"
     [ -z "$failing_checks" ] ||
         fail_condition checks-failing "checks failing: $failing_checks"
-    pending_checks="$(jq -r '.pending | join(", ")' <<<"$checks_summary")"
+    pending_checks="$(jq -r '
+        .pending | if length > 20
+        then (.[0:20] | join(", ")) + " … and \(length - 20) more"
+        else join(", ") end' <<<"$checks_summary")"
     [ -z "$pending_checks" ] ||
         fail_condition checks-pending "checks not yet concluded: $pending_checks"
 }
@@ -691,7 +710,7 @@ esac
 
 if [ "$require_draft" = 1 ]; then
     verdict_condition=ready
-    verdict_detail="every mechanically checkable readiness condition holds; ready_for_review-only automation and the Codex Auto-review knobs remain human-verified prerequisites"
+    verdict_detail="every mechanically checkable readiness condition holds"
 else
     verdict_condition=audit
     verdict_detail="every mechanically checkable condition except the draft requirement holds; this audits an existing promotion and never authorizes gh pr ready"
diff --git a/.skills-sync.yaml b/.skills-sync.yaml
index 760d69df..f6ebbd68 100644
--- a/.skills-sync.yaml
+++ b/.skills-sync.yaml
@@ -8,7 +8,7 @@
 source:
   repo: https://github.com/evanharmon1/harmon-devkit.git
   # renovate: datasource=github-releases depName=evanharmon1/harmon-devkit
-  ref: v0.29.0 # pinned tag — bump deliberately to a released harmon-devkit tag
+  ref: v0.30.0 # pinned tag — bump deliberately to a released harmon-devkit tag
 categories:
   - universal
   - repo
diff --git a/template/[% if use_skills_sync %].skills-sync.yaml[% endif %].jinja b/template/[% if use_skills_sync %].skills-sync.yaml[% endif %].jinja
index fa542aff..a8c6c34f 100644
--- a/template/[% if use_skills_sync %].skills-sync.yaml[% endif %].jinja	
+++ b/template/[% if use_skills_sync %].skills-sync.yaml[% endif %].jinja	
@@ -9,7 +9,7 @@
 source:
   repo: https://github.com/evanharmon1/harmon-devkit.git
   # renovate: datasource=github-releases depName=evanharmon1/harmon-devkit
-  ref: v0.29.0 # pinned tag — bump deliberately to a released harmon-devkit tag
+  ref: v0.30.0 # pinned tag — bump deliberately to a released harmon-devkit tag
 categories:
 [% for cat in skill_categories %]
   - [[ cat ]]