@windyroad/itil 0.59.1 → 0.59.2

package/.claude-plugin/plugin.json

@@ -497,5 +497,5 @@
     }
   },
   "name": "wr-itil",
-  "version": "0.59.1"
+  "version": "0.59.2"
 }

package/hooks/lib/changeset-detect.sh

@@ -32,10 +32,9 @@
 #       * any other path: ignored (non-publishable surface — `.github/`,
 #         root config, top-level `docs/`, etc.).
 #   - If any path is publishable source:
-#       * **Check 2a (Phase 1)**: a `.changeset/*.md` (or held-window
-#         `docs/changesets-holding/*.md` per P177) staged → allow.
-#       * **Check 2b (Phase 2)**: an in-scope `.changeset/*.md` (or
-#         held-window entry) targeting the plugin via YAML frontmatter
+#       * **Check 2a (Phase 1)**: a `.changeset/*.md` staged → allow.
+#       * **Check 2b (Phase 2)**: an in-scope `.changeset/*.md`
+#         targeting the plugin via YAML frontmatter
 #         `"@windyroad/<slug>": <any-bump>` → allow. Scope =
 #         in-unpushed-range additions (`<base>..HEAD`) + untracked
 #         working-tree files + modified-not-staged working-tree files.
@@ -98,8 +97,7 @@
 #              shape — per-invocation deterministic, no markers).
 #   P141     — this helper.
 
-# P141 Phase 2 helper — does any `.changeset/*.md` (or held entry under
-# `docs/changesets-holding/*.md`) ALREADY in scope target the plugin
+# P141 Phase 2 helper — does any `.changeset/*.md` ALREADY in scope target the plugin
 # slug via its YAML frontmatter `"@windyroad/<slug>": <bump>` line?
 #
 # Scope = files reachable from HEAD but not from `origin/<base>`,
@@ -142,11 +140,11 @@ _changeset_in_scope_covers_plugin() {
   candidates=$(
     {
       git log --diff-filter=A --name-only --pretty=format: "${base}..HEAD" \
-        -- '.changeset/*.md' 'docs/changesets-holding/*.md' 2>/dev/null
+        -- '.changeset/*.md' 2>/dev/null
       git ls-files --others --exclude-standard \
-        -- '.changeset/*.md' 'docs/changesets-holding/*.md' 2>/dev/null
+        -- '.changeset/*.md' 2>/dev/null
       git diff --name-only \
-        -- '.changeset/*.md' 'docs/changesets-holding/*.md' 2>/dev/null
+        -- '.changeset/*.md' 2>/dev/null
     } | grep -v '/README\.md$' | sort -u
   )
 
@@ -237,21 +235,6 @@ detect_changeset_required() {
       .changeset/*.md)
         has_changeset=1
         ;;
-      docs/changesets-holding/README.md)
-        # README in the holding dir is meta-doc, not a real changeset
-        # (mirrors the .changeset/README.md exclusion above).
-        ;;
-      docs/changesets-holding/*.md)
-        # P177: a held-window changeset entry IS a changeset — authored
-        # and audit-trailed, just intentionally held outside `.changeset/`
-        # per ADR-042 Rule 7 (held-window blessing). Recognising it here
-        # gives the gate a held-window-awareness branch so held-window-
-        # bound work commits no longer need a separate move-to-holding
-        # chore commit. Release/drain semantics are unchanged — the
-        # Release workflow reads `.changeset/` only; a held entry is never
-        # drained without a graduation `git mv` back into `.changeset/`.
-        has_changeset=1
-        ;;
       packages/*)
         rest="${path#packages/}"
         slug="${rest%%/*}"

package/hooks/test/itil-changeset-discipline.bats

@@ -229,54 +229,11 @@ run_bash_hook() {
   [[ "$output" == *"\"permissionDecision\": \"deny\""* ]]
 }
 
-# --- P177: held-window directory recognition (ADR-042 Rule 7) ---
-#
-# P141's gate purpose is "every publishable iter has a changeset to drain".
-# A `docs/changesets-holding/<name>.md` entry IS a changeset — authored and
-# audit-trailed, just intentionally held outside `.changeset/` per ADR-042
-# Rule 7 (held-window blessing). Before P177 the gate ignored the holding
-# directory (held entries fell through the `*)` catch-all), forcing held-
-# window-bound work through a 2-commit workaround (work commit + a separate
-# `chore(changeset): move ... to holding`). The gate now recognises a staged
-# held entry as satisfying the discipline, mirroring the `.changeset/*.md`
-# branch (and its README.md meta-doc exclusion). Release/drain semantics are
-# unchanged — the Release workflow reads `.changeset/` only; a held entry is
-# recognised at the commit-gate layer, never drained without a graduation
-# `git mv` back into `.changeset/`.
-
-@test "P177 allow: staged packages/<plugin>/ source WITH a staged docs/changesets-holding/ entry allows the commit" {
-  echo "skill body" > packages/itil/skills/foo/SKILL.md
-  mkdir -p docs/changesets-holding
-  printf -- '---\n"@windyroad/itil": patch\n---\nheld fix\n' > docs/changesets-holding/wr-itil-p177.md
-  git add packages/itil/skills/foo/SKILL.md docs/changesets-holding/wr-itil-p177.md
-  run run_bash_hook "git commit -m 'feat'"
-  [ "$status" -eq 0 ]
-  [[ "$output" != *"\"permissionDecision\": \"deny\""* ]]
-}
-
-@test "P177 allow path with held-window entry emits 0 bytes (ADR-045 Pattern 1 silent-on-pass)" {
+@test "ADR-099 deny: a staged legacy holding entry does NOT satisfy changeset discipline" {
   echo "skill body" > packages/itil/skills/foo/SKILL.md
   mkdir -p docs/changesets-holding
-  printf -- '---\n"@windyroad/itil": patch\n---\nheld fix\n' > docs/changesets-holding/wr-itil-p177.md
-  git add packages/itil/skills/foo/SKILL.md docs/changesets-holding/wr-itil-p177.md
-  run run_bash_hook "git commit -m 'feat'"
-  [ "$status" -eq 0 ]
-  [ "${#output}" -eq 0 ]
-}
-
-@test "P177 deny: staged docs/changesets-holding/README.md alone does NOT count as a valid held changeset" {
-  echo "skill body" > packages/itil/skills/foo/SKILL.md
-  mkdir -p docs/changesets-holding
-  echo "# Changesets Holding Area" > docs/changesets-holding/README.md
-  git add packages/itil/skills/foo/SKILL.md docs/changesets-holding/README.md
-  run run_bash_hook "git commit -m 'feat'"
-  [ "$status" -eq 0 ]
-  [[ "$output" == *"\"permissionDecision\": \"deny\""* ]]
-}
-
-@test "P177 deny: staged source with NEITHER .changeset/*.md NOR a holding entry still denies (regression guard)" {
-  echo "skill body" > packages/itil/skills/foo/SKILL.md
-  git add packages/itil/skills/foo/SKILL.md
+  printf -- '---\n"@windyroad/itil": patch\n---\nlegacy metadata\n' > docs/changesets-holding/legacy.md
+  git add packages/itil/skills/foo/SKILL.md docs/changesets-holding/legacy.md
   run run_bash_hook "git commit -m 'feat'"
   [ "$status" -eq 0 ]
   [[ "$output" == *"\"permissionDecision\": \"deny\""* ]]
@@ -549,20 +506,6 @@ mark_origin_at_head() {
   [[ "$output" == *"itil"* ]]
 }
 
-@test "P141 Phase 2 allow: held-window docs/changesets-holding/*.md in-range entry also covers (ADR-042 Rule 7 composes with Phase 2)" {
-  mark_origin_at_head
-  mkdir -p docs/changesets-holding
-  printf -- '---\n"@windyroad/itil": patch\n---\nheld fix\n' > docs/changesets-holding/wr-itil-p347.md
-  git add docs/changesets-holding/wr-itil-p347.md
-  git -c commit.gpgsign=false commit --quiet -m "held changeset"
-  # Subsequent itil source commit — held entry in range covers.
-  echo "skill body" > packages/itil/skills/foo/SKILL.md
-  git add packages/itil/skills/foo/SKILL.md
-  run run_bash_hook "git commit -m 'feat'"
-  [ "$status" -eq 0 ]
-  [[ "$output" != *"\"permissionDecision\": \"deny\""* ]]
-}
-
 @test "P141 Phase 2: when no upstream and no origin/main ref exists, Check 2b skips silently and Phase 1 strict-deny is preserved" {
   # No mark_origin_at_head — refs/remotes/origin/main is absent.
   # Stage source without any changeset. Phase 1 strict-deny must fire

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@windyroad/itil",
-  "version": "0.59.1",
+  "version": "0.59.2",
   "description": "ITIL-aligned IT service management for Claude Code (problem, and future incident/change skills)",
   "bin": {
     "windyroad-itil": "./bin/install.mjs"

package/scripts/derive-release-vehicle.sh

@@ -24,24 +24,12 @@
 #     merge-commit: <SHA>
 #     release-date: <YYYY-MM-DD>
 #
-# De-facto-released variant (P361, exit 0) when the changeset is present in
-# .changeset/ but its code already shipped with a sibling release:
-#   RELEASE_VEHICLE:
-#     changeset: .changeset/<name>.md
-#     status: de-facto-released (attribution pending)
-#     fix-commit: <SHA>
-#     shipped-with-version-packages-commit: <SHA>
-#     release-date: <YYYY-MM-DD>
-#     note: ...
-#
 # Exit codes:
-#   0 = OK (full citation emitted, OR de-facto-released graduated-holding
-#       changeset whose code already shipped with a sibling release — P361)
+#   0 = OK (full citation emitted)
 #   1 = ticket file not found
 #   2 = no changeset reference in ticket body AND no co-committed changeset
 #       derivable from the ticket's git history (P389 fallback)
-#   3 = changeset present in working tree AND not de-facto-released
-#       (genuinely unreleased)
+#   3 = changeset present in working tree (unreleased)
 #   4 = deletion commit found but no merge PR / merge commit resolvable
 #
 # @problem P267 — Codify derive-release-vehicle.sh helper for K→V release-
@@ -49,14 +37,6 @@
 #                 fragile to wrong-release-cited errors when sessions
 #                 pre-apply transitions across sibling tickets (observed
 #                 2026-05-18 P250 K→V cited P247's release refs).
-# @problem P361 — exit-3 "unreleased" false positive on ADR-061 graduated
-#                 holding changesets; helper now distinguishes "attribution
-#                 pending" (de-facto-released, exit 0) from "code unreleased"
-#                 (exit 3) via an add-commit ancestry test against the latest
-#                 version-packages commit.
-# @adr ADR-061 (dogfood graduation criteria — held changeset reinstated to
-#               .changeset/ awaiting attribution after its code shipped; the
-#               de-facto-released exit-0 path)
 # @adr ADR-049 (bin/ on PATH shim — adopter-safe script resolution; helper
 #               is invoked as `wr-itil-derive-release-vehicle`)
 # @adr ADR-022 (Verifying lifecycle — citation supports the K→V transition's
@@ -79,10 +59,10 @@ USAGE: derive-release-vehicle.sh <ticket-id> [<problems-dir>]
   <problems-dir> — defaults to ./docs/problems
 
 Exit codes:
-  0  ok (full citation, OR de-facto-released graduated-holding changeset — P361)
+  0  ok (full citation)
   1  ticket file not found
   2  no changeset reference in body AND no co-committed changeset in ticket history (P389)
-  3  changeset present in working tree AND not de-facto-released (unreleased)
+  3  changeset present in working tree (unreleased)
   4  deletion commit found but no merge PR / merge commit resolvable
 EOF
 }
@@ -157,57 +137,9 @@ if [ -z "$CHANGESET_PATH" ]; then
   exit 2
 fi
 
-# ── Released?  Changeset must be ABSENT from working tree (deleted by
+# ── Released? Changeset must be ABSENT from the working tree (deleted by
 #    chore: version packages) AND have a deletion commit in git history. ────
-# Exception (P361 / ADR-061 Rule 5 + P359): a changeset can be reinstated to
-# .changeset/ awaiting changelog attribution AFTER its code already de-facto
-# shipped with a sibling release (held code ships with any sibling release —
-# P359). Present-in-tree therefore does NOT always mean unreleased. Before
-# exiting 3, test whether the commit that originally ADDED this changeset is
-# an ancestor of the latest published "chore: version packages" commit; if so
-# the fix code shipped → emit a de-facto-released citation (exit 0). A
-# genuinely-unreleased fresh changeset has its add-commit NEWER than the last
-# bump, so the is-ancestor test is false and it correctly stays exit 3.
 if [ -f "$CHANGESET_PATH" ]; then
-  # Oldest Add of the path = the original fix commit. Robust to the
-  # hold→graduate `git mv`, which (without rename detection) records a later
-  # Add at the same path; `tail -1` selects the original.
-  ADD_SHA="$(
-    git log --diff-filter=A --format='%H' -- "$CHANGESET_PATH" 2>/dev/null \
-      | tail -1
-  )"
-
-  # Resolve the published-history ref (same ladder used for merge resolution).
-  DEFACTO_REF=""
-  for ref in origin/main main HEAD; do
-    if git rev-parse --verify "$ref" >/dev/null 2>&1; then
-      DEFACTO_REF="$ref"
-      break
-    fi
-  done
-
-  LATEST_VERSION_BUMP=""
-  if [ -n "$DEFACTO_REF" ]; then
-    LATEST_VERSION_BUMP="$(
-      git log --grep='^chore: version packages' --format='%H' -1 "$DEFACTO_REF" 2>/dev/null
-    )"
-  fi
-
-  if [ -n "$ADD_SHA" ] && [ -n "$LATEST_VERSION_BUMP" ] \
-     && git merge-base --is-ancestor "$ADD_SHA" "$LATEST_VERSION_BUMP" 2>/dev/null; then
-    RELEASE_DATE="$(git log -1 --format='%cs' "$LATEST_VERSION_BUMP" 2>/dev/null)"
-    cat <<EOF
-RELEASE_VEHICLE:
-  changeset: $CHANGESET_PATH
-  status: de-facto-released (attribution pending)
-  fix-commit: $ADD_SHA
-  shipped-with-version-packages-commit: $LATEST_VERSION_BUMP
-  release-date: $RELEASE_DATE
-  note: changeset present in .changeset/ awaiting changelog attribution (ADR-061 holding-graduation); code already shipped with a sibling release (P359).
-EOF
-    exit 0
-  fi
-
   echo "ERROR: changeset $CHANGESET_PATH still present in working tree (unreleased)" >&2
   exit 3
 fi

package/scripts/test/derive-release-vehicle.bats

@@ -21,11 +21,10 @@
 #     release-date: <YYYY-MM-DD>
 #
 # Exit codes:
-#   0 = OK (full citation emitted, OR de-facto-released graduated-holding
-#       changeset whose code already shipped with a sibling release — P361)
+#   0 = OK (full citation emitted)
 #   1 = ticket file not found
 #   2 = no changeset reference in ticket body
-#   3 = changeset still present AND not de-facto-released (genuinely unreleased)
+#   3 = changeset still present (unreleased)
 #   4 = deletion commit found but no merge PR / merge commit resolvable
 #
 # @adr ADR-049 (bin/ on PATH shim — adopter-safe script resolution)
@@ -158,37 +157,26 @@ EOF
   echo "$output" | grep -qi "unreleased\|not.*delet"
 }
 
-# ── Exit 0: de-facto-released graduated-holding changeset (P361) ─────────────
-
-@test "derive-release-vehicle: graduated holding changeset whose code shipped with a sibling release → exit 0 de-facto-released" {
-  # P361 / ADR-061 Rule 5 + P359: a changeset can be reinstated to
-  # .changeset/ awaiting changelog attribution AFTER its code already shipped
-  # with a sibling release. Present-in-tree must NOT read as unreleased.
-  cat > .changeset/p211-graduated.md <<'EOF'
+@test "ADR-099: present changeset remains unreleased after a sibling version bump" {
+  cat > .changeset/p211-present.md <<'EOF'
 ---
 '@windyroad/itil': patch
 ---
 
-P211 fix — held then graduated.
+P211 fix.
 EOF
-  cat > docs/problems/verifying/211-graduated.md <<'EOF'
-# Problem 211: Graduated
+  cat > docs/problems/known-error/211-present.md <<'EOF'
+# Problem 211: Present Changeset
 
 **Status**: Known Error
 
 ## Fix Strategy
 
-Ship via `.changeset/p211-graduated.md`.
+Ship via `.changeset/p211-present.md`.
 EOF
   git add .
-  git commit -q -m "feat(itil): P211 fix + changeset"   # commit A — the fix
-
-  # Hold it out of the active release queue (ADR-042 Rule 7).
-  mkdir -p docs/changesets-holding
-  git mv .changeset/p211-graduated.md docs/changesets-holding/p211-graduated.md
-  git commit -q -m "chore(itil): hold P211 changeset (above-appetite)"
+  git commit -q -m "feat(itil): P211 fix"
 
-  # A sibling release ships AFTER the fix landed (P359: code ships regardless).
   cat > .changeset/p999-sibling.md <<'EOF'
 ---
 '@windyroad/itil': patch
@@ -197,19 +185,13 @@ EOF
 Sibling fix.
 EOF
   git add .changeset/p999-sibling.md
-  git commit -q -m "feat(itil): sibling fix + changeset"
+  git commit -q -m "feat(itil): sibling fix"
   git rm -q .changeset/p999-sibling.md
-  git commit -q -m "chore: version packages"   # the published release bump
-
-  # Graduate the held changeset back to .changeset/ awaiting attribution.
-  mkdir -p .changeset
-  git mv docs/changesets-holding/p211-graduated.md .changeset/p211-graduated.md
-  git commit -q -m "chore(itil): graduate P211 changeset"
+  git commit -q -m "chore: version packages"
 
   run "$SCRIPT" P211 docs/problems
-  [ "$status" -eq 0 ]
-  echo "$output" | grep -qi "de-facto-released"
-  echo "$output" | grep -q "changeset: .changeset/p211-graduated.md"
+  [ "$status" -eq 3 ]
+  echo "$output" | grep -qi "unreleased"
 }
 
 @test "derive-release-vehicle: fresh changeset added AFTER the last release → still exit 3 (not a false de-facto-released)" {

package/skills/manage-incident/SKILL.md

@@ -168,7 +168,7 @@ Resolve each field via the following dispatch. **The order is load-bearing** —
 |-------|----------|------------------|
 | **Title** | Derive silently. Kebab-case the first 8-10 non-stopword tokens of the user's prose description (same slug derivation as `/wr-itil:capture-problem` Step 1.4 and `/wr-itil:manage-problem` Step 4). Emit stderr advisory: `manage-incident: derived title='<slug>' from description; re-invoke or rename the file if the slug is wrong`. Do NOT fire AskUserQuestion. | category-4 silent-framework |
 | **Symptoms** | Pull from user prose verbatim — the description text IS the symptoms surface for declaration. Place into the `## Observations` section template at Step 5. Do NOT fire AskUserQuestion. | category-4 silent-framework |
-| **Start time** | Derive silently, three sources in priority order: (a) explicit timestamp in description (regex `\b\d{4}-\d{2}-\d{2}([ T]\d{2}:\d{2})?\b`, or relative form `"<N> (minutes|hours|days) ago"` resolved against current wall-clock); (b) if the description cites a specific file/dir/changeset-holding-area, run `git log --diff-filter=A --follow -- <path> \| tail -1` for first-touch evidence (the I001 regression's "first hold at 2026-04-24" was this exact shape — `git log --diff-filter=A --follow -- docs/changesets-holding/`); (c) otherwise default to current wall-clock UTC. Emit stderr advisory: `manage-incident: start-time derived as <ts> from <source>; cite an additional evidence anchor in the Timeline section if symptoms began earlier`. Do NOT fire AskUserQuestion. | category-4 silent-framework |
+| **Start time** | Derive silently, three sources in priority order: (a) explicit timestamp in description (regex `\b\d{4}-\d{2}-\d{2}([ T]\d{2}:\d{2})?\b`, or relative form `"<N> (minutes|hours|days) ago"` resolved against current wall-clock); (b) if the description cites a specific file or directory, run `git log --diff-filter=A --follow -- <path> \| tail -1` for first-touch evidence; (c) otherwise default to current wall-clock UTC. Emit stderr advisory: `manage-incident: start-time derived as <ts> from <source>; cite an additional evidence anchor in the Timeline section if symptoms began earlier`. Do NOT fire AskUserQuestion. | category-4 silent-framework |
 | **Severity** | Derive silently when evidence maps to a clear `RISK-POLICY.md` Impact × Likelihood cell. Cross-reference description signals against the matrix: (a) impact signals (service disruption keywords like `down` / `degraded` / `unavailable` → high; latency / throughput keywords → moderate; cosmetic / typo keywords → low); (b) likelihood signals (`reproducible` / `every request` → high; `intermittent` / `flaky` → medium; `one-off` / `single user` → low); (c) named anchors (held-cluster age cited → use that age to map cell; scorer state cited → use the cited band). When the cross-reference produces a single clear cell, set it silently and emit stderr advisory: `manage-incident: severity derived as <score> (<label>) from RISK-POLICY matrix + evidence: <evidence list>; re-invoke or update if mis-rated`. **Ambiguous-evidence fallback** (no mappable signal in description, or signals point to conflicting cells): fire AskUserQuestion with the Impact (1-5) × Likelihood (1-5) options as the genuine ADR-044 **category-5 (taste)** fallback surface. The fallback is genuine ambiguity, NOT defaults. | category-4 silent-framework (derivable); category-5 fallback (ambiguous) |
 | **Scope** | Retain AskUserQuestion. Scope is the user-judgment surface — only the user knows whether downstream-adopter-risk is in scope, whether mobile is affected, whether the blast radius extends past the cited symptoms. The framework cannot resolve semantic scope deterministically (same reasoning as Step 2 duplicate-check). Construct the call with `header: "Incident scope"`, `multiSelect: false` if a closed enum applies or free-text capture otherwise. This is the canonical ADR-044 **category-1 (direction-setting)** surface — *"only the user knows the goals that haven't been written down yet."* | category-1 direction-setting |
 

package/skills/manage-problem/SKILL.md

@@ -1084,7 +1084,7 @@ Commit the completed work per ADR-014 (governance skills commit their own work):
    - Fix implemented: `fix(<scope>): <description> (closes P<NNN>)` — include problem file changes (rename to `.verifying.md` + `## Fix Released` section) in the same commit per ADR-022
 5. If commit risk is above appetite: auto-apply scorer remediations per **ADR-042 Rule 1** incrementally until residual commit risk is within appetite (≤ 4/25), OR halt per ADR-042 Rule 5 if the scorer cannot converge. **The skill MUST NOT commit above appetite, and MUST NOT call `AskUserQuestion` to ask whether to commit anyway** (P377/RFC-029 amendment 2026-06-24 — above-appetite is framework-mediated, never a category-3 one-time-override; same invariant the push/release branch at Step 12 already enforces). The ADR-013 Rule 6 fail-safe (no `AskUserQuestion` available / non-interactive → skip the commit and report the uncommitted state) remains the terminal fallback. This applies only to the risk-above-appetite branch, not to the delegation-unavailable case above.
 
-**Multi-commit slice changeset discipline (P141 Phase 2)**: when a single logical fix lands across multiple ADR-014-grain commits targeting the same plugin (e.g. helper extraction in commit 1, callers wired in commit 2, SKILL note + transition in commit 3 — all `packages/<plugin>/`), author ONE changeset on the first commit in the slice. Subsequent same-plugin commits do NOT need their own changeset — the `itil-changeset-discipline.sh` hook's Check 2b recognises any `.changeset/*.md` (or held-window `docs/changesets-holding/*.md` entry) already in the unpushed slice scope (`origin/<base>..HEAD` + untracked + modified-not-staged) that targets `"@windyroad/<plugin>": <any-bump>` and allows. This eliminates the per-commit changeset ceremony that previously produced N redundant `.changeset/*.md` files for one logical release entry (changesets-action collapses bump-class at version-package time, so per-commit changesets rendered N near-identical CHANGELOG bullets for one release). Once a changeset hits `origin/<base>` (drained at release time), it no longer counts — a fresh changeset is required for the next slice. Cross-plugin coverage is NOT permitted: an `@windyroad/itil` changeset does not satisfy a `packages/voice-tone/` commit.
+**Multi-commit slice changeset discipline (P141 Phase 2)**: when a single logical fix lands across multiple ADR-014-grain commits targeting the same plugin (e.g. helper extraction in commit 1, callers wired in commit 2, SKILL note + transition in commit 3 — all `packages/<plugin>/`), author ONE changeset on the first commit in the slice. Subsequent same-plugin commits do NOT need their own changeset — the `itil-changeset-discipline.sh` hook's Check 2b recognises any `.changeset/*.md` already in the unpushed slice scope (`origin/<base>..HEAD` + untracked + modified-not-staged) that targets `"@windyroad/<plugin>": <any-bump>` and allows. This eliminates the per-commit changeset ceremony that previously produced N redundant `.changeset/*.md` files for one logical release entry (changesets-action collapses bump-class at version-package time, so per-commit changesets rendered N near-identical CHANGELOG bullets for one release). Once a changeset hits `origin/<base>` (drained at release time), it no longer counts — a fresh changeset is required for the next slice. Cross-plugin coverage is NOT permitted: an `@windyroad/itil` changeset does not satisfy a `packages/voice-tone/` commit.
 
 ### 12. Auto-release when changesets are queued (ADR-020)
 
@@ -1121,8 +1121,9 @@ Otherwise, after the commit in step 11 lands, drain the release queue so the fix
 2. Read the descriptions. Decide what to do. The agent MAY follow a scorer suggestion, adapt it, or do something else entirely. There is no requirement to rank all suggestions upfront or iterate through them in order.
 3. **Verification Pending carve-out (ADR-042 Rule 2b)**: skip remediations that target a commit attached to a `.verifying.md` ticket. Do NOT auto-revert VP commits.
 4. Apply the chosen action using standard primitives (git, Edit, Bash). Example actions:
-   - `move-to-holding`: `git mv .changeset/<name>.md docs/changesets-holding/<name>.md` + append to holding-area README "Currently held" per ADR-042 Rule 6. Since the non-AFK skill has no iteration wrapper to amend into, each auto-apply is its own commit (ADR-042 Rule 3). Each commit goes through the standard ADR-014 commit flow — architect + JTBD + risk-scorer gates.
    - `revert-commit`: `git revert --no-edit <sha>`. The scorer SHOULD supply the target commit SHA in the `description` column. Before executing, verify the SHA is NOT attached to a `.verifying.md` ticket (Rule 2b carve-out). After revert, commit the revert as a standalone auto-apply commit (no amend folding in non-AFK mode). If `git revert` produces merge conflicts, route to Rule 5 halt with the conflict detail.
+   - `split-change`: split the work into independently releasable, within-appetite commits and re-score each slice.
+   - `disable-or-revert`: remove the risky behaviour from shipped code until its required evidence or control exists.
 5. Re-score via the same delegation path as step 1 above.
 6. **Loop**: re-score within appetite → drain per the Drain action above. Re-score still above → continue working to reduce risk. The agent reads the new remediations and decides what to do next. Loop. Exhausted or unsupported class → Rule 5 halt.
 
@@ -1130,7 +1131,7 @@ Otherwise, after the commit in step 11 lands, drain the release queue so the fix
 - The final `RISK_SCORES:` line
 - An "Auto-apply trail" subsection listing each remediation attempted with outcome
 - Any Verification Pending ticket IDs implicated per Rule 2b
-- A one-line scorer-gap note (e.g., "scorer produced only `move-to-holding`; residual still ≥ 5/25 after exhaustion — extend scorer vocabulary per P108")
+- A one-line scorer-gap note (e.g., "available remediations did not reduce residual risk below 5/25 — extend scorer vocabulary per P108")
 
 The user resolves interactively — typical resolutions include splitting the commit, feature-flagging the change, or opening a problem ticket documenting the scorer gap.
 

package/skills/manage-rfc/SKILL.md

@@ -259,11 +259,7 @@ Satisfy the commit gate per ADR-014 — primary path delegates to `wr-risk-score
 
 ### 12. Auto-release (skip in AFK orchestrator)
 
-Same conditional drain as `manage-problem` Step 12: if not in an AFK orchestrator AND `.changeset/` is non-empty AND push/release within appetite, run `npm run push:watch` then `npm run release:watch`. Held-changeset window per ADR-042 / P162 governs which changesets graduate; RFC-shaped held changesets graduate atomically per ADR-060 architect finding 12.
-
-## Held-changeset window scope (Phase 1)
-
-Phase 1 of the RFC framework (Slices 2-5 per `docs/plans/170-rfc-framework-story-map.md`) ships under a held-changeset window. ADR-042 auto-apply is paused until RFC-001 (P168 retro) reaches `closed` status. Counterfactual risk assessment per P162 governs graduation: delay-risk vs release-risk. The full chain graduates atomically — the entire RFC-001 commit chain ships or nothing does.
+Same conditional drain as `manage-problem` Step 12: if not in an AFK orchestrator AND `.changeset/` is non-empty AND push/release are within appetite, run `npm run push:watch` then `npm run release:watch`.
 
 ## Composition with capture-rfc
 
@@ -291,7 +287,7 @@ The two skills share the `/tmp/wr-itil-rfc-capture-grep-${SESSION_ID}` create-ga
 - **ADR-022** — lifecycle suffix-based (RFC mirrors).
 - **ADR-032** — lightweight + heavyweight split.
 - **ADR-038** — progressive disclosure; future REFERENCE.md split deferred per ADR-054.
-- **ADR-042** — held-changeset auto-apply; window discipline.
+- **ADR-042** — above-appetite remediation and halt discipline.
 - **ADR-044** — decision delegation contract; authority classes.
 - **ADR-049** — `wr-itil-reconcile-rfcs` shim grammar (Slice 3).
 - **ADR-051** — load-bearing-from-the-start; I1 hard-block on day one.
@@ -304,4 +300,4 @@ The two skills share the `/tmp/wr-itil-rfc-capture-grep-${SESSION_ID}` create-ga
 - **P134** — Last-reviewed line discipline (single fragment + history archive).
 - **P138** — tie-break ladder consistency.
 - **P150** — Verification Queue sort direction.
-- **P162** — held-changeset graduation criteria.
+- **P359** — changeset holding does not withhold shipped code.

package/skills/update-upstream/SKILL.md

@@ -18,7 +18,7 @@ allowed-tools: Read, Write, Edit, Bash, Glob, Grep, AskUserQuestion, Skill, Agen
   @adr ADR-010 amended (sibling-skill naming; split execution ownership)
   @adr ADR-044 (decision-delegation contract — framework-resolution boundary)
   @adr ADR-075 (Amendment 2026-06-02 — paired promptfoo eval discharges R009 prose floor)
-  @adr ADR-061 Rule 4 (evidence-floor — paired Tier-A/B eval ships in same commit as SKILL)
+  @adr ADR-075 (evidence-floor — paired Tier-A/B eval ships in same commit as SKILL)
 -->
 
 # Update Upstream — Lifecycle-Update Skill
@@ -549,7 +549,7 @@ The skill's no-op exit (Step 1) means firing the trigger unconditionally on ever
 - [ADR-044](../../../docs/decisions/044-decision-delegation-contract.proposed.md) — framework-resolution boundary; the gate verdict IS the framework, no per-transition AskUserQuestion for the gate firing itself.
 - [ADR-042](../../../docs/decisions/042-auto-apply-scorer-remediations-open-vocabulary.proposed.md) — within-axis precedent for open-vocabulary risk-reducing measures.
 - [ADR-075](../../../docs/decisions/075-promptfoo-agent-prose-verdict-eval-harness.proposed.md) Amendment 2026-06-02 — paired promptfoo Tier-A/B eval discharges the R009 prose-floor for SKILL surfaces.
-- [ADR-061](../../../docs/decisions/061-dogfood-graduation-criteria.proposed.md) Rule 4 — evidence-floor; the paired eval ships in the same commit as this SKILL prose for atomic R009 discharge.
+- [ADR-075](../../../docs/decisions/075-agent-prose-behavioural-test-harness.proposed.md) — evidence-floor; the paired eval ships in the same commit as this SKILL prose for atomic R009 discharge.
 - **P080** — driving problem ticket (No bidirectional update of upstream-reported problems).
 - **P363** — driving problem ticket for the [§ Inbound-origin verdict dispatch](#inbound-origin-verdict-dispatch-p363) leg (inbound-reported tickets never received a fix-released verdict on the originating issue). Fix option (b) — consume the `**Origin**` field — user-ratified 2026-06-22. **Rework 2026-06-23** (four user directives): the inbound verdict is LLM-generated per-context (not templated); O→KE shares the workaround with reporter/commenter provenance-credit; anti-leakage is visibility-gated (PUBLIC → titled+linked refs); the gate chain gains a cog-a11y leg first (P338-gated).
 - [ADR-055](../../../docs/decisions/055-plugin-published-namespace-prefixed-internal-ids.proposed.md) — grounds the PUBLIC-repo titled+linked reference discipline (permalink-progressive-enhancement; cures the bare-ID collision failure mode).

package/skills/update-upstream/test/update-upstream-contract.bats

@@ -346,12 +346,11 @@ setup() {
   [ "$status" -eq 0 ]
 }
 
-# ─── Paired promptfoo eval (ADR-075 Amendment 2026-06-02 + ADR-061 Rule 4) ────
+# ─── Paired promptfoo eval (ADR-075 Amendment 2026-06-02) ────────────────────
 #
 # The paired promptfoo eval at packages/itil/skills/update-upstream/eval/
 # discharges the R009 prose-floor for this SKILL surface atomically per
-# ADR-061 Rule 4 evidence-floor — without the paired eval, ADR-042 Rule 2
-# move-to-holding would apply per the P080 iter contract.
+# ADR-075 evidence floor.
 
 @test "update-upstream: paired promptfoo eval config exists (ADR-075 Amendment 2026-06-02)" {
   [ -f "$EVAL_CONFIG" ]

package/skills/work-problems/SKILL.md

@@ -367,7 +367,6 @@ Before the orchestrator emits the final `ALL_DONE` sentinel for the AFK loop, it
    - it is `verifying` / carries `## Fix Released` awaiting user verification (stop-condition #2, interactive);
    - it carries an upstream-blocked marker (`## Reported Upstream` / `- **Upstream report pending** --` / em-dash legacy) or a recorded blocked classification with a dead-end investigation (stop-condition #3);
    - it was filtered out THIS session by Step 3.5 (interactive-ratification predicate) or Step 3.6 (already-shipped relevance gate) — keyed off the durable per-session skip record those steps write (the `outstanding_questions` entry in `.afk-run-state/outstanding-questions.jsonl` carrying the ticket id), NOT agent recollection, so the classification is reproducible across the re-scan and cannot loop forever;
-   - its fix changeset sits in `docs/changesets-holding/` with an unmet reinstate criterion (held, not dispatchable).
    Every other open / known-error ticket is **dispatchable** — ordinary autonomous fix-and-commit work. The agent MUST NOT reclassify a dispatchable ticket as non-dispatchable because the *salient* remainder of the backlog is interactive-gated, because the ticket "feels" out of scope, or because a user-directed pivot consumed the loop's attention. The subjective "this is a natural stopping point" judgement is exactly the P390 failure; the classification is per-ticket and marker-bound. **The classification MUST be PRINTED as a table in the turn output** (ticket → dispatchable/non-dispatchable → the recorded marker that decided) — not merely computed. A computed-but-unprinted re-scan is invisible to the Step 0e `/goal` external evaluator, which judges only what the transcript surfaces (ADR-026 grounding); the printed table is the evidence the canonical goal condition names.
 
 3. *Decide.* If the re-scan yields **≥1 dispatchable ticket**, `ALL_DONE` is FORBIDDEN: the stop-condition the orchestrator was about to emit does NOT objectively hold. The orchestrator loops back to Step 3 tier-first selection (Critical-bypass → Inbound-reported → Internal, within-tier WSJF per ADR-076) over the dispatchable set and dispatches the next iter — it does NOT proceed to gate (a)/(b)/(c). Only when the re-scan yields **zero dispatchable tickets** does gate (0) pass and the sequence proceed to gate (a). Gate (0) finding work is a **loopback, not a halt** — it is productive (the loop resumes draining), so it is NOT a Hard-fail halt trigger.
@@ -889,32 +888,8 @@ After the iteration's commit lands but before starting the next iteration, check
 2. Read the returned `RISK_SCORES: commit=X push=Y release=Z` line and the `RISK_REMEDIATIONS:` block (if present).
 3. **Classify the residual + queue state (P250)**:
    - **Above appetite (≥ 5/25)** — route to the **Above-appetite branch** below. Do NOT drain. Do NOT proceed to Step 6.75 until either (a) the auto-apply loop re-converges within appetite and drain succeeds, or (b) Rule 5 halt fires.
-   - **Within appetite (≤ 4/25) AND there is releasable material** (any unpushed commits on `HEAD..origin/<base>` OR any entries in `.changeset/` OR any graduation-eligible entries in `docs/changesets-holding/` per ADR-061 Rule 1 that are not VP-blocked per Rule 2) — drain the queue per the Drain action below, then proceed to Step 6.75. The release-action threshold is "is there something to release?", NOT "has accumulated risk reached the safety band?" Per user direction 2026-05-17 (P250 Description): *"If it's low risk, you should release."* Low cost to release + low residual risk = release now; never accumulate.
-   - **Within appetite (≤ 4/25) AND empty queue** (no unpushed commits AND no `.changeset/` entries AND no graduation-eligible held entries) — no drain (literally nothing to release). Proceed to Step 6.75. This is the genuine no-op fast-path; the gate is *absence of releasable material*, not residual band.
-
-**Cohort-graduation pre-check (per ADR-061 Rule 5; P246):** when the within-appetite-with-releasable-material branch fires AND `docs/changesets-holding/` is non-empty, invoke the graduation evaluator BEFORE the Drain action. The evaluator is the deterministic Rule 1a join + Rule 2 VP carve-out + Rule 3b cohort-grouping pass shipped in `@windyroad/risk-scorer` Phase 2a/2b. **The graduation criterion is evidence-of-working-as-desired (Rule 4 per-class evidence floor), not elapsed wall-clock time** — per user direction 2026-05-17: *"Dogfooding makes sense, but it shouldn't be time based, it should be until we are happy that it's working as desired."* + *"Why are we waiting? That seems to go against the principles if you ask me."* Calendar predicates are NEVER a primary graduation trigger; the evaluator's `status=resolved` IS the graduation signal.
-
-1. Run the shim: `wr-risk-scorer-evaluate-graduation` (resolves to `packages/risk-scorer/scripts/evaluate-graduation.sh` per ADR-049 naming grammar). The script enumerates `docs/changesets-holding/*.md` (excluding README), applies ADR-061 Rule 1a join + Rule 2 VP carve-out + Rule 3b cohort grouping, and emits one `GRADUATION_CANDIDATE:` line per held entry plus a final `GRADUATION_SUMMARY:` line.
-
-   **Evaluator scope (load-bearing per P308)**: the evaluator script implements ONLY the deterministic Rule 1a + Rule 2 + Rule 3b passes. It does NOT compute release-risk and does NOT apply Rule 4 evidence-floor judgement — those are LLM-judgement surfaces owned by the orchestrator + the `wr-risk-scorer:pipeline` agent. Therefore evaluator `status=resolved` means *"the ticket-join succeeded and the entry is not VP-blocked"* only — it is **necessary but not sufficient** for graduation. The Rule 4 evidence-floor judgement (per-class evidence per ADR-061 Rule 4) MUST run as a separate orchestrator-side step before any `git mv` (P308 amendment).
-2. Parse each `GRADUATION_CANDIDATE: changeset=<basename> | ticket=<P-id> | priority=<N> | class=<3a|3b> | [cohort=<id> |] status=<resolved|vp-blocked|halt-no-resolution>` line. Branch on `status`:
-   - **`status=resolved`** — route to **Rule 4 evidence-floor judgement** (P308 amendment; see step 2a below). The evaluator's `status=resolved` is necessary-but-not-sufficient; Rule 4 judgement is the LLM-owned surface that ratifies the *evidence floor* per ADR-061 Rule 4 + ADR-044 framework-resolution boundary. Do NOT auto-graduate at this point. Per ADR-061 Rule 5: the graduation criterion authorises the *intent*; Rule 4 judgement is the precondition that admits *evaluation* of that criterion.
-   - **`status=vp-blocked`** — skip. Per ADR-061 Rule 2 (Verification Pending carve-out; symmetric to ADR-042 Rule 2b). Do NOT graduate; held entry stays. The `.verifying.md` → `.closed.md` transition auto-clears the carve-out at a later Step 6.5 graduation pass.
-   - **`status=halt-no-resolution`** — halt. Per ADR-061 Rule 1a terminal: when neither filename-convention join nor body-grep fallback resolves a ticket, OR the resolved ticket file is missing/unreadable, the orchestrator MUST NOT auto-graduate. Route to the **Step 6.5 cohort-graduation halt-no-resolution** halt point (framework-prescribed halt — see Mid-loop ask discipline subsection); halt-with-batched-questions per the Step 2.5b cross-reference.
-
-2a. **Rule 4 evidence-floor judgement (P308 — LLM-owned, NOT framework-resolved)**. For each `status=resolved` candidate, the orchestrator MUST evaluate whether the class-specific evidence floor has been met before graduating. The evidence floor (ADR-061 Rule 4) is per-class: PreToolUse:Bash gates need ≥1 gate-fire trace; UserPromptSubmit detectors need ≥1 detector firing; commit-hook-with-auto-fix needs ≥1 correctness-verified auto-fix commit; SessionStart additionalContext needs ≥1 session-trail entry. The orchestrator reads the held entry's `Reinstate criterion (evidence-based, ...)` line from `docs/changesets-holding/README.md` Currently held entry + cross-references the on-disk evidence artefact named in the criterion. Per ADR-044, Rule 4 judgement is LLM-owned (not framework-resolved). Route per AskUserQuestion availability:
-
-    - **Interactive (`AskUserQuestion` available)**: fire `AskUserQuestion` **per held entry** (NOT batched across cohort — each entry's evidence is independent, except for class=3b cohorts which graduate atomically per Rule 3b — when any cohort member's `AskUserQuestion` returns Graduate, ALL same-cohort members graduate together; when any returns Defer/Reject, the entire cohort stays held). Question text MUST inline (P350 brief-before-ID discipline): the held entry's basename + ticket + Priority, the per-entry evidence summary (from the README's Currently held entry — Rule 4 class-specific shape + cited artefact), and the candidate verdict. Options: **Graduate (evidence cited and met)** / **Defer (evidence not yet met — preserve hold)** / **Reject (manual intervention — surface back to user)**. Per ADR-013 Rule 1 cap (≤4 per call, sequential if >4). On `Graduate`, fall through to step 2b graduate-action. On `Defer`/`Reject`, the held entry stays in `docs/changesets-holding/`; emit one Auto-apply-trail line per ADR-061 Rule 6 citing the user's verdict + reason; proceed to next candidate. This per-entry AskUserQuestion is a framework-prescribed user-interaction surface — see Mid-loop ask discipline subsection.
-    - **AFK (`AskUserQuestion` forbidden — iter subprocess / non-interactive context per P352 / ADR-013 Rule 6)**: queue one `outstanding_question` entry (category: `direction`) per held entry to `.afk-run-state/outstanding-questions.jsonl`. The entry's `question` field inlines the same evidence summary the interactive `AskUserQuestion` would surface (P350 brief-before-ID — do NOT use opaque IDs alone). Do NOT graduate. Continue Drain action: any pre-existing `.changeset/` entries still drain per the within-appetite contract (those entries were not graduations and are unaffected). Per P352 user-ratified universal AFK default: queue-and-continue (NOT halt, NOT silent skip, NOT auto-default). Loop-end Step 2.5 surfaces the queued questions as batched `AskUserQuestion` per the existing accumulated-question discipline.
-
-2b. **Graduate-action (Rule 4-ratified path)**. When Rule 4 evidence-floor judgement returns `Graduate` (interactive `AskUserQuestion` Graduate verdict) for a held entry: perform `git mv docs/changesets-holding/<basename> .changeset/<basename>`. Append the entry to `docs/changesets-holding/README.md` "Recently reinstated" with the citation: `<basename> — graduation criterion met (status=resolved per Rule 1a join to <P-id>, Priority <N>; Rule 4 evidence-floor ratified per user verdict <verbatim>); class <3a|3b>; evidence cited`. For class=3b cohorts: when any cohort member's Rule 4 judgement returns `Graduate`, ALL same-cohort members with `status=resolved` graduate together atomically (Rule 3b cohort propagation — entire cohort ships or none does); when any returns Defer/Reject, the entire cohort stays held. Amend the iter's main commit per ADR-042 Rule 3 amend-based folding to preserve the ADR-032 one-commit-per-iteration invariant.
-
-3. After processing all candidates: if anything graduated, the just-moved entries are now in `.changeset/` and ride the existing Drain action (no separate re-entry needed — the Drain action's `release:watch` step picks them up when `.changeset/` is non-empty). Proceed to the Drain action below.
-4. **Governance gates apply (ADR-061 Rule 7)**: every graduation reinstate goes through the standard ADR-014 commit flow — architect / JTBD / risk-scorer gates ride the amend commit; gate rejection routes to ADR-042 Rule 5 halt with the rejection reason logged. The graduation criterion authorises the *intent*; the gates authorise the *action*.
-
-**Idempotency**: safe to invoke when holding-area is empty (script exits 1 with `GRADUATION_SUMMARY: total=0` — orchestrator skips graduation, proceeds to Drain action). Safe when no candidates resolve (all `vp-blocked`) — no `git mv` operations, no README mutation, no commit amendment. Safe when AFK + all `status=resolved` route to queue (no graduation performed; outstanding_questions accumulate for loop-end surfacing per P352 / ADR-013 Rule 6).
-
-**Audit trail (ADR-061 Rule 6)**: every `reinstate-from-holding` graduation appends one Auto-apply-trail line to the iter report AND one "Recently reinstated" line to `docs/changesets-holding/README.md` with the resolved problem-ticket ID, Priority value, graduation class (3a or 3b), and the evidence citation. The audit trail is the load-bearing artefact for ADR-026 cite + persist + uncertainty grounding.
+   - **Within appetite (≤ 4/25) AND there is releasable material** (any unpushed commits on `HEAD..origin/<base>` OR any entries in `.changeset/`) — drain the queue per the Drain action below, then proceed to Step 6.75. The release-action threshold is "is there something to release?", NOT "has accumulated risk reached the safety band?" Per user direction 2026-05-17 (P250 Description): *"If it's low risk, you should release."* Low cost to release + low residual risk = release now; never accumulate.
+   - **Within appetite (≤ 4/25) AND empty queue** (no unpushed commits AND no `.changeset/` entries) — no drain (literally nothing to release). Proceed to Step 6.75. This is the genuine no-op fast-path; the gate is *absence of releasable material*, not residual band.
 
 **Drain action (non-interactive, policy-authorised per ADR-013 Rule 6):**
 
@@ -948,8 +923,6 @@ ADR-022 prescribes that Known Error tickets transition to Verification Pending o
 
 **Composition with the Above-appetite branch (below)**: the K→V callback is anchored to the within-appetite Drain action step 4 — it does NOT fire after the above-appetite Rule 5 halt (no release shipped → nothing to match) and it does NOT fire mid-loop in the above-appetite auto-apply loop. When the auto-apply loop converges and re-enters the within-appetite Drain action, the K→V callback fires there per step 4.
 
-**Composition with Cohort-graduation pre-check (P246)**: the cohort-graduation pre-check (step 2a above) fires BEFORE the Drain action; its `git mv` operations from `docs/changesets-holding/` to `.changeset/` happen BEFORE release:watch and ship as part of the same release. The K→V callback fires AFTER release:watch and consumes the just-shipped changeset set — so graduated cohorts that ship in the same release are correctly matched by the enumerator (the deleted-from-tree changeset has the graduated basename; the K-ticket's `**Release vehicle**: .changeset/<basename>.md` reference matches).
-
 Per ADR-022 (Verifying lifecycle) + ADR-018 (release-cadence host) + ADR-010 amended P093 (transition-problem authoritative executor) + ADR-014 (per-transition commit grain) + ADR-013 Rule 5 (policy-authorised silent-proceed) + ADR-044 (framework-resolution boundary) + P228 (this ticket) + P233 (sibling callback) + P267 (derive-release-vehicle composed helper) + P330 (Release vehicle seed reference — input signal).
 
 **Post-release cache refresh (P233) — fires only after within-appetite Drain action step 5 (above):**
@@ -1006,11 +979,12 @@ After a successful release-cadence drain has shipped a new plugin version to npm
    - R1 | <description> | <effort S/M/L> | <risk_delta -N> | <files affected>
    - R2 | ...
    ```
-2. Read the descriptions. Decide what to do. The agent MAY follow a scorer suggestion, adapt it, or do something else entirely. There is no requirement to rank all suggestions upfront or iterate through them in order.
+2. Read the descriptions. Decide what to do. The agent MAY follow a scorer suggestion, adapt it, or do something else entirely. There is no requirement to rank all suggestions upfront or iterate through them in order. The selected action and the branch report MUST name how actual shipped risk changes (for example: split the change, add a control, disable the risky behaviour, or revert it); generic "apply remediation" wording is not enough.
 3. **Verification Pending carve-out (ADR-042 Rule 2b)**: if a remediation targets a commit attached to a `.verifying.md` ticket, do NOT auto-revert it. Skip that suggestion and decide on the next one.
 4. Apply the chosen action using standard primitives (git, Edit, Bash). Example actions the agent might take:
-   - `move-to-holding`: `git mv .changeset/<name>.md docs/changesets-holding/<name>.md`. Append the entry to `docs/changesets-holding/README.md` under "Currently held" per ADR-042 Rule 6. Amend the iteration's commit to fold the move (per ADR-042 Rule 3 amend-based folding — preserves ADR-032 one-commit-per-iteration invariant).
    - `revert-commit`: `git revert --no-edit <sha>`. The scorer SHOULD supply the target commit SHA in the `description` column (e.g., "Revert commit 9a1f96c that introduced the risky gate"). Before executing, verify the SHA is NOT attached to a `.verifying.md` ticket (Rule 2b carve-out). After revert, amend the iteration's commit to fold the revert. If `git revert` produces merge conflicts, route to Rule 5 halt with the conflict detail — do not attempt non-interactive conflict resolution.
+   - `split-change`: split the iteration into independently releasable, within-appetite commits and re-score each slice.
+   - `disable-or-revert`: remove the risky behaviour from shipped code until its required evidence or control exists.
 5. Re-invoke the risk scorer (same delegation path as step 1 above — subagent preferred, skill fallback). Read the new `RISK_SCORES:` line.
 6. **Loop classification**:
    - **Re-score within appetite (≤ 4/25)** — proceed to Drain action above. Done with the above-appetite branch.
@@ -1025,13 +999,13 @@ After a successful release-cadence drain has shipped a new plugin version to npm
 - The final `RISK_SCORES:` line
 - An "Auto-apply trail" subsection listing each remediation attempted with outcome
 - Any Verification Pending ticket IDs implicated per Rule 2b
-- A one-line scorer-gap note (e.g., "scorer produced only `move-to-holding` remediations; residual still ≥ 5/25 after exhaustion — extend scorer vocabulary per P108")
+- A one-line scorer-gap note (e.g., "available remediations did not reduce residual risk below 5/25 — extend scorer vocabulary per P108")
 
 **Step 2.5b cross-reference (P126)**: before emitting the Rule 5 halt iteration summary, run Step 2.5b's surfacing routine. The routine is gated on ≥1 accumulated user-answerable skip; Rule 5 halts that fire late in a long AFK loop frequently have accumulated skips from prior iters, so Step 2.5b's AskUserQuestion-default branch typically fires (`halt-paths-must-route-design-questions-through-Step-2.5b`). **Critical guard (architect FLAG)**: Step 2.5b surfaces *prior-iter accumulated user-answerable skips only* — it does NOT ask the user how to remediate the above-appetite state itself; the halt-causing scorer-gap remains a halt-with-bug-signal per ADR-042 Rule 5 invariant ("never release above appetite", scorer is the decision surface, not the user). Surfacing prior-iter skips does not retry the above-appetite remediation, does not bypass the never-release-above-appetite invariant, and does not convert the halt into a non-halt — it just takes the existing prior-iter user-input round-trip with it.
 
-Halt is a **bug signal** — the scorer should always have progressively more aggressive remediations available once P108 lands. Until then, exhaustion is expected when the only path to within-appetite requires a non-`move-to-holding` class.
+Halt is a **bug signal** — the scorer should always have progressively more aggressive remediations available once P108 lands.
 
-**Audit trail (ADR-042 Rule 6):** append one line per auto-apply to the iteration summary's Auto-apply trail subsection, including remediation ID, action class, pre/post scores, action taken, and description citation. For `move-to-holding` actions, also append to `docs/changesets-holding/README.md` "Currently held".
+**Audit trail (ADR-042 Rule 6):** append one line per auto-apply to the iteration summary's Auto-apply trail subsection, including remediation ID, action class, pre/post scores, action taken, and description citation.
 
 ### Step 6.75: Inter-iteration verification (P036)
 
@@ -1086,11 +1060,8 @@ When `AskUserQuestion` is unavailable or the user is AFK, the skill (and the del
 | Scope expansion during work | Update problem file, re-score WSJF, move to next problem instead of continuing |
 | Commit when risk within appetite | Auto-commit (manage-problem step 9e fallback) |
 | Commit when risk above appetite | Skip commit, report uncommitted state |
-| Pipeline risk within appetite (≤ 4/25) with releasable material (any unpushed commits OR any `.changeset/` entries OR any graduation-eligible held entries per ADR-061 Rule 1) | Drain release queue (`push:watch` then, if releasable changesets exist, `release:watch`) before next iteration — per ADR-018 (Step 6.5) as amended by P250. Trigger is *presence of releasable material*, not residual band reaching appetite. User direction 2026-05-17: "If it's low risk, you should release." |
-| Pipeline risk within appetite (≤ 4/25) AND empty queue (no unpushed commits AND no `.changeset/` AND no graduation-eligible held entries) | No drain — literally nothing to release. Proceed directly to Step 6.75. The genuine no-op fast-path per P250. |
-| Cohort-graduation pre-check fires before Drain action (within-appetite branch, `docs/changesets-holding/` non-empty) — evaluator returns `status=resolved` | Route to Rule 4 evidence-floor judgement (LLM-owned per ADR-061 Rule 4 + ADR-044 framework-resolution boundary). Evaluator's `status=resolved` is necessary-but-not-sufficient (P308 — evaluator script disclaims Rule 4 at lines 19-22). Interactive: per-held-entry `AskUserQuestion` with inline evidence summary (P350 brief-before-ID) + 3 options (Graduate / Defer / Reject). AFK: queue per-held-entry `outstanding_question` to `.afk-run-state/outstanding-questions.jsonl` (P352 / ADR-013 Rule 6 queue-and-continue universal default) — do NOT graduate, continue Drain for any pre-existing `.changeset/` entries. On Graduate verdict: `git mv docs/changesets-holding/<basename> .changeset/<basename>`, append README "Recently reinstated" entry citing the user's Rule 4 verdict, amend the iter's main commit per ADR-042 Rule 3. For class=3b cohorts, all cohort members graduate atomically on any-member Graduate verdict (Rule 3b cohort propagation); any Defer/Reject keeps entire cohort held. Per ADR-061 Rule 4 + Rule 5 + Rule 6 + Rule 7 + ADR-013 Rule 6 + P246 + P308 + P350 + P352 (Step 6.5 Cohort-graduation pre-check; step 2a Rule 4 evidence-floor judgement). Graduation criterion is evidence-of-working-as-desired (Rule 4 evidence floor), not elapsed wall-clock time — user direction 2026-05-17: "Dogfooding makes sense, but it shouldn't be time based, it should be until we are happy that it's working as desired." |
-| Cohort-graduation pre-check — evaluator returns `status=vp-blocked` | Skip. Per ADR-061 Rule 2 Verification Pending carve-out (symmetric to ADR-042 Rule 2b). Do NOT graduate; held entry stays. `.verifying.md` → `.closed.md` transition auto-clears the carve-out at a later pass. Per ADR-061 Rule 2 + P246. |
-| Cohort-graduation pre-check — evaluator returns `status=halt-no-resolution` | Halt at the framework-prescribed "Step 6.5 cohort-graduation halt-no-resolution" halt point. Per ADR-061 Rule 1a terminal: ambiguous join is a user-decision surface, not an agent-decision surface. Halt-with-batched-questions per the Step 2.5b cross-reference. Per ADR-061 Rule 1a + P246. |
+| Pipeline risk within appetite (≤ 4/25) with releasable material (any unpushed commits OR any `.changeset/` entries) | Drain release queue (`push:watch` then, if releasable changesets exist, `release:watch`) before next iteration — per ADR-018 (Step 6.5) as amended by P250. Trigger is *presence of releasable material*, not residual band reaching appetite. User direction 2026-05-17: "If it's low risk, you should release." |
+| Pipeline risk within appetite (≤ 4/25) AND empty queue (no unpushed commits AND no `.changeset/`) | No drain — literally nothing to release. Proceed directly to Step 6.75. The genuine no-op fast-path per P250. |
 | Post-release K→V auto-transition between iters (P228) | After a successful within-appetite Drain action shipped a release to npm, invoke `wr-itil-enumerate-postrelease-kv-candidates` to enumerate `.known-error.md` tickets whose `**Release vehicle**: .changeset/<name>.md` citation matches a just-shipped (deleted-from-tree) changeset, and dispatch `/wr-itil:transition-problem <NNN> verifying` per emitted `KV_CANDIDATE` line. Conditional on actual release (skipped when `push:watch` ran alone with no changeset); non-blocking on individual transition failure (logs per-ticket, continues to next candidate; persistent failures route to Step 2.5b accumulated questions). V→C remains a maintainer-only surface — this callback fires K→V only. Per ADR-022 + ADR-018 + ADR-010 amended P093 + ADR-014 + ADR-013 Rule 5 + ADR-044 + P228 + P233 + P267 + P330 (Step 6.5 Post-release K→V auto-transition subsection). |
 | Post-release plugin cache refresh between iters (P233) | After a successful within-appetite Drain action shipped a release to npm, chain `/install-updates` to refresh the plugin cache before the next iter dispatches. Conditional on actual release (skipped when `push:watch` ran alone with no changeset); non-blocking on `/install-updates` failure (degrades to cache-stays-stale, equivalent to pre-amendment behaviour). Mid-loop ask discipline preserved by treating any `/install-updates` AskUserQuestion surface AS the Non-interactive fallback dry-run path. Per ADR-013 Rule 5 + ADR-044 + P130 + P106 + P233 (Step 6.5 Post-release cache refresh subsection). |
 | CI failure during Step 6.5 drain (within-appetite branch) | Diagnose via `gh run view --log-failed`, classify against the closed fixable-in-iter allow-list (P081-class stale-grep-string, hook stub mismatch, test ID drift, environmental flake), fix-and-continue for fixable classes (each retry rides its own ADR-014 commit gate), 3-retry cap per iteration, halt for unrecoverable classes. Ambiguous classification defaults to halt. ADR-013 Rule 5 policy-authorised. Per ADR-026 grounding + ADR-044 framework-resolution boundary + P140 (Step 6.5 Failure handling). |
@@ -1105,7 +1076,7 @@ When `AskUserQuestion` is unavailable or the user is AFK, the skill (and the del
 | Unexpected dirty state between iterations | Halt the loop. Report the `git status --porcelain` output, the last iteration's reported outcome, and the divergence — per P036 (Step 6.75). Run Step 2.5b before emitting the halt summary if ≥1 accumulated user-answerable skip from prior iters (P126). Do NOT attempt non-interactive recovery of the dirty state itself. |
 | Iter committed cleanly + claim contradicts on-disk ADR Confirmation state (P335) | Halt the loop with `outcome: halted-iter-over-claim`. Include the `wr-itil-verify-iter-summary` stdout (the `OVER-CLAIM: ADR-NNN has N unchecked Confirmation item(s)...` lines) as the divergence detail. Run Step 2.5b before emitting the halt summary if ≥1 accumulated user-answerable skip from prior iters. Do NOT auto-correct the iter's claim — the orchestrator cannot retroactively make a false claim true; the user adjudicates on return (re-dispatch / accept partial / amend). Per ADR-013 Rule 6 + ADR-032 subprocess-boundary trust contract + P335 (Step 6.75 verify-iter-claims sub-step). |
 | External root cause detected at Open → Known Error, or at park with `upstream-blocked` reason | **Auto-invoke `/wr-itil:report-upstream`** via the manage-problem Step 6 external-root-cause detection AFK fallback (per ADR-024 2026-06-04 (P270) amendment). The report-upstream skill composes the draft then scores the prose via `wr-risk-scorer:external-comms` (ADR-028); below-appetite → sends (public-issue Step 5 / comment Step 5c / security Step 6 per classification); above-appetite → risk-reduces (open-ended LLM judgement per leaf (a)) then re-scores → sends-or-queues to `## Queued Upstream Report` (leaf (c)). Security routing per leaf (b): upstream-with-`SECURITY.md` + below-appetite → files via declared channel; upstream-without-`SECURITY.md` → external-comms-gated impact assessment. Queue does NOT halt (P352). Tickets already carrying the stable `- **Upstream report pending** -- external dependency identified; invoke /wr-itil:report-upstream when ready` marker from prior sessions are detected via the already-noted grep check and routed to the report-upstream invocation; the marker shape is retained as the detection substrate (ASCII `--` per P210 — em-dash variant is the legacy form, still matched by the already-noted check for backward compatibility). Per P063 (amended 2026-06-04) + P270 + ADR-013 Rule 6. |
-| Mid-loop ask between iters in the orchestrator's main turn | Forbidden except at framework-prescribed user-interaction points (Step 0 session-continuity / fetch-failure halt; Step 2.5 / 2.5b loop-end emit; Step 6.5 above-appetite Rule 5 halt; Step 6.5 CI-failure / release:watch halt; Step 6.5 cohort-graduation halt-no-resolution halt; Step 6.5 cohort-graduation per-entry Rule 4 evidence-floor judgement (P308 — interactive only; AFK queues per P352); Step 6.75 dirty-for-unknown-reason halt). The loop's purpose is **progress + accumulation**; mechanical-stage transitions between iters are framework-resolved and MUST NOT prompt the user. Per ADR-044 framework-resolution boundary + ADR-013 Rule 1 (as amended by ADR-044) + P130. |
+| Mid-loop ask between iters in the orchestrator's main turn | Forbidden except at framework-prescribed user-interaction points (Step 0 session-continuity / fetch-failure halt; Step 2.5 / 2.5b loop-end emit; Step 6.5 above-appetite Rule 5 halt; Step 6.5 CI-failure / release:watch halt; Step 6.75 dirty-for-unknown-reason halt). The loop's purpose is **progress + accumulation**; mechanical-stage transitions between iters are framework-resolved and MUST NOT prompt the user. Per ADR-044 framework-resolution boundary + ADR-013 Rule 1 (as amended by ADR-044) + P130. |
 
 ### Mid-loop ask discipline (orchestrator main turn) — P130
 
@@ -1118,8 +1089,6 @@ The orchestrator MUST NOT call `AskUserQuestion` between iterations except at th
 - **Step 2.5 / Step 2.5b loop-end emit** — accumulated `outstanding_questions` queue presented as batched `AskUserQuestion` (or fallback Outstanding Design Questions table per ADR-013 Rule 6). This is the framework's prescribed user-interaction point; do NOT dilute it by asking earlier.
 - **Step 6.5 above-appetite Rule 5 halt** — auto-apply loop exhausted without convergence; halt-with-batched-questions per the Step 2.5b cross-reference (Step 2.5b surfaces *prior-iter accumulated user-answerable skips only* — the halt-causing scorer-gap remains a halt-with-bug-signal per ADR-042 Rule 5).
 - **Step 6.5 CI-failure / `release:watch` failure halt** — push:watch or release:watch failed AND the failure is genuinely-unrecoverable (outside the fixable-in-iter allow-list, or 3-retry cap reached); halt-with-batched-questions per the Step 2.5b cross-reference. Failures inside the closed allow-list route to fix-and-continue per Step 6.5 Failure handling (P140), not this halt point.
-- **Step 6.5 cohort-graduation halt-no-resolution halt (P246)** — graduation evaluator returned `status=halt-no-resolution` for one or more held candidates (Rule 1a terminal: neither filename-convention join nor body-grep fallback resolved a problem ticket, OR the resolved ticket file is missing/unreadable). The orchestrator MUST NOT auto-graduate under ambiguity per ADR-061 Rule 1a; halt-with-batched-questions per the Step 2.5b cross-reference. The halt-causing ambiguity itself remains a halt-with-bug-signal (the held entry stays in `docs/changesets-holding/`; manual reinstate or ticket-file correction required); Step 2.5b surfaces *prior-iter accumulated user-answerable skips only* and does NOT ask the user to resolve the ambiguity itself.
-- **Step 6.5 cohort-graduation per-entry Rule 4 evidence-floor judgement (P308) — interactive only** — graduation evaluator returned `status=resolved` for ≥1 held candidate AND `AskUserQuestion` is available. Per ADR-061 Rule 4 + ADR-044 framework-resolution boundary, Rule 4 evidence-floor judgement is LLM-owned (not framework-resolved); the user ratifies per held entry with Graduate / Defer / Reject before any `git mv`. This is NOT a halt — the orchestrator continues the loop after the user verdict (graduate path performs git mv + README append + ADR-042 Rule 3 amend; defer/reject paths preserve the hold). When `AskUserQuestion` is unavailable (AFK path), the orchestrator queues `outstanding_question` entries per held candidate per P352 / ADR-013 Rule 6 queue-and-continue universal default — does NOT halt, does NOT silently proceed, does NOT auto-default. The held entries' user ratifications then surface at Step 2.5 loop-end via the existing accumulated-questions discipline.
 - **Step 6.75 dirty-for-unknown-reason halt** — `git status --porcelain` divergence; halt-with-batched-questions per the Step 2.5b cross-reference.
 - **Step 6.75 iter-over-claim halt (P335)** — `wr-itil-verify-iter-summary` detected the iter's commit message or `ITERATION_SUMMARY.notes` contains completion-claim language for an ADR whose `## Confirmation` section still has unchecked `- [ ]` items; halt-with-batched-questions per the Step 2.5b cross-reference. The over-claim itself remains a halt-with-bug-signal — Step 2.5b surfaces *prior-iter accumulated user-answerable skips only*; it does NOT ask the user how to remediate the false claim (re-dispatch / accept partial / amend the commit remains a user decision on return).
 
@@ -1228,7 +1197,7 @@ When every skipped ticket is in the `upstream-blocked` category (stop-condition
 ## Related
 
 - **P341** (`docs/problems/open/341-work-problems-skill-must-surface-outstanding-questions-then-run-retro-before-emitting-all-done.md`) — driver for Step 2.4 Pre-`ALL_DONE` gate sequence (UNCONDITIONAL fire of outstanding-questions surface + session-level retro before `ALL_DONE` emit). 2026-05-31 user direction (verbatim in ticket Description): *"The work-problems skill MUST surface the outstanding questions at the end before emitting ALL_DONE. It MUST then run a retro. Only then should it emit ALL_DONE."* Closes the structural gap that allowed `ALL_DONE` to emit while direction-class observations remained queued AND without a session-level retro running. Behavioural second-source: `test/work-problems-p341-pre-all-done-gate.bats`. Composes with P086 (extends iter-level retro-on-exit to orchestrator-level), P126 (preserves `halt-paths-must-route-design-questions-through-Step-2.5b` principle), ADR-014 (retro commits its own work), ADR-044 (framework-resolution boundary for when to surface — now framework-resolved as unconditional pre-`ALL_DONE`).
-- **P390** (`docs/problems/known-error/390-agent-declares-all-done-prematurely-while-actionable-backlog-remains.md`) — driver for Step 2.4 **Gate (0) — Objective backlog-empty assertion** (prepended ahead of gate (a)). Bug shape: the orchestrator emitted `ALL_DONE` while a dispatchable Tier-2 backlog remained, by generalising "the *salient* remainder is interactive-gated" to "Step 2 stop-condition #2 holds" — a subjective stop the framework did not authorise; it also skipped P382 (Tier-1 sev-16) entirely. Fix: before `ALL_DONE`, gate (0) re-scans the live open/known-error backlog (fresh dual-tolerant glob, not the Step 1 cache) and classifies each ticket dispatchable/non-dispatchable OBJECTIVELY by recorded marker (verifying / `## Fix Released`; upstream-blocked; blocked dead-end; Step 3.5/3.6 durable per-session skip record; held changeset). ≥1 dispatchable ticket FORBIDS `ALL_DONE` and loops back to Step 3 tier-first selection (loopback, not halt); a user-directed pivot does not discharge the Tier-exhaustion obligation. Sibling loop-control-stop class: P332 (run-retro skip rationalisation), P148 (Stage-1 ticketing skip), P175 (scope-pin loop-control inference); hardens P341's precondition. Behavioural second-source: `eval/promptfooconfig.yaml` Tier-A regex + Tier-B llm-rubric asserting the orchestrator does NOT emit `ALL_DONE` when ≥1 dispatchable ticket remains. Per ADR-044 "Continue / stop loops" framework-resolution (the natural stop is concrete — `ALL_DONE` conditions objectively met). **Reopened fix (2026-07-05 → 2026-07-06, ADR-094 / RFC-047 / STORY-040)**: Gate (0) alone proved insufficient (self-assessment — the same actor decides "should I stop" and "is stopping justified"); Step 0e anchors the loop with the native `/goal` external evaluator (per-turn small-fast-model judgement of the printed gate (0) table), gate (0)'s classification MUST be printed in turn output as the evaluator's evidence, and the headless launch shape `claude -p "/goal <canonical condition>"` is the anchor-guaranteed path (interactive starts get a nudge-and-proceed fallback).
+- **P390** (`docs/problems/known-error/390-agent-declares-all-done-prematurely-while-actionable-backlog-remains.md`) — driver for Step 2.4 **Gate (0) — Objective backlog-empty assertion** (prepended ahead of gate (a)). Bug shape: the orchestrator emitted `ALL_DONE` while a dispatchable Tier-2 backlog remained, by generalising "the *salient* remainder is interactive-gated" to "Step 2 stop-condition #2 holds" — a subjective stop the framework did not authorise; it also skipped P382 (Tier-1 sev-16) entirely. Fix: before `ALL_DONE`, gate (0) re-scans the live open/known-error backlog (fresh dual-tolerant glob, not the Step 1 cache) and classifies each ticket dispatchable/non-dispatchable OBJECTIVELY by recorded marker (verifying / `## Fix Released`; upstream-blocked; blocked dead-end; Step 3.5/3.6 durable per-session skip record). ≥1 dispatchable ticket FORBIDS `ALL_DONE` and loops back to Step 3 tier-first selection (loopback, not halt); a user-directed pivot does not discharge the Tier-exhaustion obligation. Sibling loop-control-stop class: P332 (run-retro skip rationalisation), P148 (Stage-1 ticketing skip), P175 (scope-pin loop-control inference); hardens P341's precondition. Behavioural second-source: `eval/promptfooconfig.yaml` Tier-A regex + Tier-B llm-rubric asserting the orchestrator does NOT emit `ALL_DONE` when ≥1 dispatchable ticket remains. Per ADR-044 "Continue / stop loops" framework-resolution (the natural stop is concrete — `ALL_DONE` conditions objectively met). **Reopened fix (2026-07-05 → 2026-07-06, ADR-094 / RFC-047 / STORY-040)**: Gate (0) alone proved insufficient (self-assessment — the same actor decides "should I stop" and "is stopping justified"); Step 0e anchors the loop with the native `/goal` external evaluator (per-turn small-fast-model judgement of the printed gate (0) table), gate (0)'s classification MUST be printed in turn output as the evaluator's evidence, and the headless launch shape `claude -p "/goal <canonical condition>"` is the anchor-guaranteed path (interactive starts get a nudge-and-proceed fallback).
 - **P342** (`docs/problems/open/342-iter-retros-queue-observations-as-outstanding-questions-instead-of-auto-ticketing-same-trust-boundary-as-step-4a.md`) — driver for Step 5 iter-prompt body's retro-surfaced observation classification taxonomy and capture-* carve-out. Iter retros' observations of recurring class-of-behaviour now route to `/wr-itil:capture-problem` (mechanical-stage carve-out per run-retro Step 4a precedent); only direction-setting observations queue at `outstanding_questions`; ambiguous defaults to auto-ticket per the trust-boundary asymmetry. The "no `capture-*` siblings mid-loop" rule is preserved for non-retro mid-iter capture (P078-class spam); the carve-out is bounded to the retro path. Sibling locus: `packages/retrospective/skills/run-retro/SKILL.md` Step 4b carries the symmetric mirror (same trust-boundary fires whether retro runs in iter context OR standalone in main turn). Behavioural second-source: `test/work-problems-p342-retro-auto-ticket-carveout.bats` + `packages/retrospective/skills/run-retro/test/run-retro-step-4b-retro-auto-ticket-carveout.bats`. Composes with run-retro Step 4a (precedent), ADR-013 Rule 5 (policy-authorised silent proceed), ADR-032 (foreground-spawns-N-background fanout already documented for Stage 1 in run-retro Step 4b), ADR-044 (mechanical-stage carve-out), P130 (mid-loop AskUserQuestion ban unchanged), P078 (capture-on-correction — distinct trigger surface; both end in capture but for different signals).
 - **P121** (`docs/problems/121-afk-orchestrator-should-sigterm-stuck-subprocesses-after-idle-timeout.verifying.md`) — driver for Step 5's backgrounded-poll-loop dispatch shape (replacing the prior foreground-synchronous form) and the idle-timeout SIGTERM branch. The 2026-04-25 P118 iter 5 evidence: an iteration subprocess sat idle ~70 min after its final commit, then SIGTERM produced a clean JSON exit-flush. Fix: orchestrator backgrounds the subprocess, polls every 60s, computes `LAST_ACTIVITY_MARK = max(DISPATCH_START_EPOCH, git log -1 --format=%at HEAD)`, and sends SIGTERM when `now - LAST_ACTIVITY_MARK > WORK_PROBLEMS_IDLE_TIMEOUT_S` (default 3600s = 60 min). Behavioural second-source: `test/work-problems-step-5-idle-timeout-sigterm.bats` exercises a fake `claude -p` shim that sleeps past the threshold and asserts SIGTERM, JSON exit-flush, env-var override, and within-threshold no-fire. Step 6's per-iter progress line SHOULD annotate `(SIGTERM_SENT)` when the branch fires so users can distinguish recovered iters from natural completions. ADR-032's subprocess-boundary variant amended 2026-04-26 with the backgrounded-poll-loop refinement.
 - **P146** (`docs/problems/146-afk-iteration-subprocess-bash-until-loop-polls-bats-output-with-bats-console-regex-against-tap-format.verifying.md`) — driver for Step 5 iteration prompt body's bats-output-polling-discipline clause. The 2026-04-29 incident (iter 1, PID 23580 child PID 16408) saw a `bash until`-loop poll a backgrounded bats output file with regex `^[0-9]+ tests?,` (bats's *default* console-summary format) against `bats --tap` output that never emits that line — silent infinite spin after bats completed; manual SIGTERM at 68m34s wall-clock; metadata loss per the P147 stuck-before-emit subclass. The polling idiom is NOT taught by any SKILL.md (audit confirmed via repo grep) — it is agent-learned from training data. Fix: prompt-discipline rule in the iteration prompt body's Constraints list explicitly forbidding the antipattern, naming `wait $bg_pid` (or Bash-tool `run_in_background=true` + `BashOutput`) as the safe substitute, and citing the TAP-vs-console-summary divergence so future contributors don't "fix" the rule incorrectly. Behavioural second-source: `test/work-problems-step-5-bats-polling-discipline.bats` asserts the prohibition phrase, the safe-substitute pointer, the P146 cite, the divergence explanation, and the Related-section cite.

package/skills/work-problems/test/work-problems-above-appetite-remediation.bats

@@ -46,17 +46,18 @@ setup() {
   [ "$status" -eq 0 ]
 }
 
-@test "SKILL.md references docs/changesets-holding/ (Rule 2a move-to-holding class)" {
-  # The one currently-implemented action class moves changesets to the holding
-  # area. The path must be named so the skill body is unambiguous about target.
+@test "SKILL.md does not use a held-changeset directory as remediation" {
   run grep -n "docs/changesets-holding/" "$SKILL_FILE"
-  [ "$status" -eq 0 ]
+  [ "$status" -ne 0 ]
 }
 
-@test "SKILL.md names the open action-class vocabulary (Rule 2a)" {
-  # "move-to-holding" is the single supported class today; later P108 extends.
-  # The string must appear so the enumeration is greppable.
+@test "SKILL.md does not offer move-to-holding as a remediation" {
   run grep -n "move-to-holding" "$SKILL_FILE"
+  [ "$status" -ne 0 ]
+}
+
+@test "SKILL.md offers shipment-affecting remediation classes" {
+  run grep -nE "split-change|disable-or-revert|revert-commit" "$SKILL_FILE"
   [ "$status" -eq 0 ]
 }
 

package/skills/work-problems/test/work-problems-step-6-5-always-drain.bats

@@ -4,8 +4,7 @@
 # pivot on releasable material, not residual band. The defective prior
 # clause "Within appetite (≤ 3/25) — no drain needed" encoded
 # accumulation-permitted-below-threshold semantics that violated the
-# symmetric-balance principle (ADR-061 Rule 1) and the user's verbatim
-# direction: "If it's low risk, you should release."
+# user's verbatim direction: "If it's low risk, you should release."
 #
 # Amended contract (three-band):
 #   1. Above appetite (≥ 5/25)        → ADR-042 auto-apply (unchanged).
@@ -13,8 +12,7 @@
 #   3. Within appetite (≤ 4/25) AND empty queue        → no drain (no-op fast-path).
 #
 # The trigger for the drain action is *presence of releasable material*
-# (any unpushed commits OR any .changeset/ entries OR any graduation-
-# eligible held entries per ADR-061 Rule 1). The residual band remains
+# (any unpushed commits OR any .changeset/ entries). The residual band remains
 # the safety check (above-appetite never releases) but is no longer the
 # action gate for the within-appetite branch.
 #
@@ -32,7 +30,6 @@
 # @adr ADR-018 (release-cadence policy parent — amended in same commit)
 # @adr ADR-037 (skill-testing strategy — contract-assertion class)
 # @adr ADR-042 (above-appetite branch — preserved unchanged)
-# @adr ADR-061 (Rule 1 symmetric-balance principle — parent principle)
 # @jtbd JTBD-006 (Progress the Backlog While I'm Away — primary)
 # @jtbd JTBD-002 (Ship with Confidence — composes; small frequent releases)
 
@@ -85,7 +82,7 @@ setup() {
 
 @test "work-problems P250: Within-appetite + releasable material triggers drain" {
   # The load-bearing positive contract: within appetite AND any unpushed
-  # commits OR changeset OR graduation-eligible held entry → drain.
+  # commits OR changeset entry → drain.
   run grep -nE 'Within appetite \(≤ 4/25\) AND there is releasable material' "$SKILL_MD"
   [ "$status" -eq 0 ]
 }
@@ -100,14 +97,6 @@ setup() {
   [ "$status" -eq 0 ]
 }
 
-@test "work-problems P250: releasable-material clause enumerates ADR-061 Rule 1 graduation-eligible held entries" {
-  # ADR-061 cross-reference: the symmetric-balance disjunct ensures
-  # held entries that have decayed within appetite are graduation-
-  # eligible AND drainable.
-  run grep -nE 'graduation-eligible.*ADR-061 Rule 1|ADR-061 Rule 1.*graduation-eligible|docs/changesets-holding.*ADR-061' "$SKILL_MD"
-  [ "$status" -eq 0 ]
-}
-
 # ── Within-appetite + empty queue no-op fast-path ─────────────────────────
 
 @test "work-problems P250: Within-appetite + empty queue does NOT drain (no-op fast-path)" {

package/skills/work-problems/test/work-problems-step-6-5-cohort-graduation.bats

@@ -1,326 +0,0 @@
-#!/usr/bin/env bats
-
-# P246: /wr-itil:work-problems Step 6.5 cohort-graduation pre-check must
-# invoke the deterministic graduation evaluator BEFORE the Drain action
-# when the within-appetite-with-releasable-material branch fires AND
-# docs/changesets-holding/ is non-empty.
-#
-# Refined framing (per user direction 2026-05-17): graduation criterion
-# is positive evidence that the surface works as desired (ADR-061 Rule 4
-# per-class evidence floor), NOT elapsed wall-clock time. Calendar
-# predicates (`≥7 days in-repo dogfood`, `on or after <date>`) are NEVER
-# a primary graduation trigger.
-#
-# Pre-check parses `GRADUATION_CANDIDATE` lines from the evaluator and
-# branches per the 3-status taxonomy the evaluator actually emits:
-#
-#   status=resolved              → git mv from holding to .changeset/,
-#                                   README "Recently reinstated" append,
-#                                   amend iter commit per ADR-042 Rule 3
-#                                   (policy-authorised silent proceed per
-#                                    ADR-013 Rule 5 + ADR-061 Rule 5).
-#   status=vp-blocked            → skip (ADR-061 Rule 2 VP carve-out).
-#   status=halt-no-resolution    → halt at the framework-prescribed
-#                                   "Step 6.5 cohort-graduation halt-
-#                                   no-resolution" halt point (ADR-061
-#                                   Rule 1a terminal).
-#
-# Class=3b cohorts graduate atomically per ADR-061 Rule 3b — entire
-# cohort ships or none does.
-#
-# Doc-lint contract assertions per ADR-037 Permitted Exception (contract-
-# assertion class). The asserted prose IS the load-bearing policy surface
-# — re-reading SKILL.md is the only way an AFK reader (and the iteration
-# subprocess) learns the new pre-check behaviour. These tests function as
-# regression guards against re-introducing calendar-trigger framings,
-# silently removing the pre-check, or drifting the 3-status branching.
-#
-# @problem P246
-# @adr ADR-061 (parent principle — Rules 1/1a/2/3/4/5/6/7)
-# @adr ADR-042 (Rule 3 amend-based folding for graduation reinstate commit)
-# @adr ADR-018 (release-cadence policy parent — drain condition unchanged)
-# @adr ADR-013 (Rule 5 policy-authorised silent proceed)
-# @adr ADR-037 (skill-testing strategy — contract-assertion class)
-# @adr ADR-044 (framework-resolution boundary — no AskUserQuestion mid-iter)
-# @jtbd JTBD-006 (Progress the Backlog While I'm Away — primary)
-# @jtbd JTBD-302 (Trust That the README Describes the Plugin I Just Installed)
-# @jtbd JTBD-001 (Enforce Governance Without Slowing Down)
-# @jtbd JTBD-101 (Extend the Suite with New Plugins)
-
-setup() {
-  REPO_ROOT="$(cd "$(dirname "$BATS_TEST_FILENAME")/../../../../.." && pwd)"
-  SKILL_MD="$REPO_ROOT/packages/itil/skills/work-problems/SKILL.md"
-  HOLDING_README="$REPO_ROOT/docs/changesets-holding/README.md"
-}
-
-# ── Preconditions ──────────────────────────────────────────────────────────
-
-@test "work-problems P246: SKILL.md exists" {
-  [ -f "$SKILL_MD" ]
-}
-
-@test "work-problems P246: changesets-holding README exists" {
-  [ -f "$HOLDING_README" ]
-}
-
-# ── Pre-check sub-step exists and cites the evaluator shim ────────────────
-
-@test "work-problems P246: SKILL.md Step 6.5 contains 'Cohort-graduation pre-check' sub-step" {
-  run grep -nE 'Cohort-graduation pre-check.*ADR-061 Rule 5.*P246' "$SKILL_MD"
-  [ "$status" -eq 0 ]
-}
-
-@test "work-problems P246: pre-check invokes wr-risk-scorer-evaluate-graduation shim" {
-  run grep -nE 'wr-risk-scorer-evaluate-graduation' "$SKILL_MD"
-  [ "$status" -eq 0 ]
-}
-
-@test "work-problems P246: pre-check fires BEFORE the Drain action (on within-appetite branch)" {
-  # The pre-check must precede the Drain action heading; if SKILL.md
-  # ever inverts the order, the just-graduated entries would not ride
-  # the existing release flow.
-  pre_check_line=$(grep -nE '^\*\*Cohort-graduation pre-check' "$SKILL_MD" | head -1 | cut -d: -f1)
-  drain_line=$(grep -nE '^\*\*Drain action ' "$SKILL_MD" | head -1 | cut -d: -f1)
-  [ -n "$pre_check_line" ]
-  [ -n "$drain_line" ]
-  [ "$pre_check_line" -lt "$drain_line" ]
-}
-
-# ── Three-status branching contract ───────────────────────────────────────
-
-@test "work-problems P246/P308: status=resolved routes to Rule 4 evidence-floor judgement (NOT auto-graduate)" {
-  # P308 amendment: status=resolved is necessary-but-not-sufficient.
-  # The evaluator script disclaims Rule 4 (LLM-owned). SKILL prose must
-  # surface Rule 4 evidence-floor judgement before any git mv.
-  run grep -nE '`status=resolved`.*route to.*Rule 4 evidence-floor judgement' "$SKILL_MD"
-  [ "$status" -eq 0 ]
-}
-
-@test "work-problems P308: Rule 4-ratified graduate path performs git mv from holding to .changeset/" {
-  # The git mv is downstream of the Rule 4 evidence-floor ratification,
-  # not directly on the evaluator's status=resolved.
-  run grep -nE 'git mv docs/changesets-holding/<basename> \.changeset/<basename>' "$SKILL_MD"
-  [ "$status" -eq 0 ]
-}
-
-@test "work-problems P308: Rule 4-ratified graduate path amends the iter commit per ADR-042 Rule 3" {
-  run grep -nE 'Amend the iter.s main commit per ADR-042 Rule 3 amend-based folding' "$SKILL_MD"
-  [ "$status" -eq 0 ]
-}
-
-@test "work-problems P246: pre-check branches on status=vp-blocked → skip (ADR-061 Rule 2)" {
-  run grep -nE '`status=vp-blocked`.*skip.*ADR-061 Rule 2' "$SKILL_MD"
-  [ "$status" -eq 0 ]
-}
-
-@test "work-problems P246: pre-check branches on status=halt-no-resolution → halt at framework-prescribed point" {
-  run grep -nE '`status=halt-no-resolution`.*halt.*ADR-061 Rule 1a terminal' "$SKILL_MD"
-  [ "$status" -eq 0 ]
-}
-
-@test "work-problems P246: SKILL.md does NOT branch on non-existent status=no-graduate-evidence-floor" {
-  # Architect verdict: the evaluator emits only resolved | vp-blocked |
-  # halt-no-resolution. A SKILL branch handling a non-existent token
-  # would be dead code (Confirmation Violation).
-  run grep -nE 'no-graduate-evidence-floor' "$SKILL_MD"
-  [ "$status" -ne 0 ]
-}
-
-# ── Cohort propagation (class=3b atomic) ──────────────────────────────────
-
-@test "work-problems P246: class=3b cohorts graduate atomically (entire cohort ships or none does)" {
-  run grep -nE 'class=3b cohorts.*ALL members.*graduate together atomically|cohort ships or none does' "$SKILL_MD"
-  [ "$status" -eq 0 ]
-}
-
-@test "work-problems P246: cohort propagation cites ADR-061 Rule 3b" {
-  run grep -nE 'Rule 3b cohort propagation' "$SKILL_MD"
-  [ "$status" -eq 0 ]
-}
-
-# ── Framework-prescribed halt point added ─────────────────────────────────
-
-@test "work-problems P246: new halt point 'Step 6.5 cohort-graduation halt-no-resolution' in Mid-loop ask discipline list" {
-  run grep -nE 'Step 6\.5 cohort-graduation halt-no-resolution halt.*P246' "$SKILL_MD"
-  [ "$status" -eq 0 ]
-}
-
-@test "work-problems P246: halt-no-resolution halt routes through Step 2.5b cross-reference" {
-  # Architect verdict: halt-no-resolution must inherit the established
-  # Step 2.5b routing pattern (CI failure / Rule 5 above-appetite /
-  # dirty-unknown all route this way).
-  run grep -nE 'cohort-graduation.*Step 2\.5b cross-reference|halt-with-batched-questions per the Step 2\.5b cross-reference' "$SKILL_MD"
-  [ "$status" -eq 0 ]
-}
-
-@test "work-problems P246: Mid-loop ask between iters row enumerates the new halt point" {
-  run grep -nE 'Mid-loop ask between iters.*Step 6\.5 cohort-graduation halt-no-resolution halt' "$SKILL_MD"
-  [ "$status" -eq 0 ]
-}
-
-# ── Non-Interactive Decision Making table reflects amendment ──────────────
-
-@test "work-problems P246: Decision Making table carries the resolved-graduate row" {
-  run grep -nE '\| Cohort-graduation pre-check.*status=resolved' "$SKILL_MD"
-  [ "$status" -eq 0 ]
-}
-
-@test "work-problems P246: Decision Making table carries the vp-blocked-skip row" {
-  run grep -nE '\| Cohort-graduation pre-check.*status=vp-blocked' "$SKILL_MD"
-  [ "$status" -eq 0 ]
-}
-
-@test "work-problems P246: Decision Making table carries the halt-no-resolution-halt row" {
-  run grep -nE '\| Cohort-graduation pre-check.*status=halt-no-resolution' "$SKILL_MD"
-  [ "$status" -eq 0 ]
-}
-
-# ── Evidence-based criterion (calendar-trigger rejection) ─────────────────
-
-@test "work-problems P246: pre-check states graduation criterion is evidence-of-working-as-desired, NOT elapsed wall-clock time" {
-  run grep -nE 'evidence-of-working-as-desired.*not elapsed wall-clock time|not elapsed wall-clock time' "$SKILL_MD"
-  [ "$status" -eq 0 ]
-}
-
-@test "work-problems P246: pre-check explicitly rejects calendar predicates as primary triggers" {
-  run grep -nE 'Calendar predicates are NEVER a primary graduation trigger' "$SKILL_MD"
-  [ "$status" -eq 0 ]
-}
-
-@test "work-problems P246: pre-check cites user direction verbatim ('Dogfooding makes sense, but it shouldn't be time based')" {
-  run grep -nE "Dogfooding makes sense, but it shouldn.t be time based, it should be until we are happy that it.s working as desired" "$SKILL_MD"
-  [ "$status" -eq 0 ]
-}
-
-@test "work-problems P246: pre-check cites user correction verbatim ('Why are we waiting?')" {
-  run grep -nE "Why are we waiting\?.*That seems to go against the principles if you ask me" "$SKILL_MD"
-  [ "$status" -eq 0 ]
-}
-
-# ── Policy authorisation (ADR-013 Rules 1/5/6 + ADR-061 Rules 4/5 + P352) ─
-
-@test "work-problems P308: pre-check cites ADR-061 Rule 4 as the LLM-owned evidence-floor surface" {
-  # Rule 4 evidence-floor judgement is class-specific and LLM-owned per
-  # ADR-044 framework-resolution boundary; surfacing it as AskUserQuestion
-  # (interactive) or queue (AFK) is the load-bearing delegation surface.
-  run grep -nE 'ADR-061 Rule 4' "$SKILL_MD"
-  [ "$status" -eq 0 ]
-}
-
-@test "work-problems P308: pre-check cites ADR-013 Rule 6 as the AFK queue-and-continue fail-safe" {
-  # Per P352 user-ratified universal default: when AskUserQuestion is
-  # unavailable (AFK iter), queue the question + continue — do NOT halt,
-  # do NOT silently fail-soft, do NOT auto-default.
-  run grep -nE 'ADR-013 Rule 6.*P352|P352.*ADR-013 Rule 6' "$SKILL_MD"
-  [ "$status" -eq 0 ]
-}
-
-@test "work-problems P308: pre-check cites ADR-061 Rule 5 (graduation criterion authorises intent, gates authorise action)" {
-  run grep -nE 'ADR-061 Rule 5' "$SKILL_MD"
-  [ "$status" -eq 0 ]
-}
-
-@test "work-problems P308: interactive branch fires AskUserQuestion per-held-entry (not silent proceed)" {
-  # The amendment lifts the prior silent-proceed clause — Rule 4 judgement
-  # is LLM-owned; the user must ratify before the git mv.
-  run grep -nE 'interactive.*AskUserQuestion.*per[- ]held[- ]entry|per[- ]held[- ]entry.*AskUserQuestion' "$SKILL_MD"
-  [ "$status" -eq 0 ]
-}
-
-@test "work-problems P308: AFK branch queues outstanding_question per held entry (queue-and-continue)" {
-  run grep -nE 'AFK.*queue.*outstanding_question.*per held entry|queue.*outstanding_question.*per held entry.*AFK|AFK.*queue.*outstanding_question|outstanding_question.*do NOT graduate' "$SKILL_MD"
-  [ "$status" -eq 0 ]
-}
-
-@test "work-problems P308: AskUserQuestion option labels inline evidence + Rule 4 criterion (P350 brief-before-ID)" {
-  # P350 empathy-gap correction: AskUserQuestion text must inline the
-  # per-held-entry evidence summary + Rule 4 class-specific criterion;
-  # opaque P-ID / ADR-ID references alone are non-compliant.
-  run grep -nE 'P350|brief.*before.*ID|inline.*evidence' "$SKILL_MD"
-  [ "$status" -eq 0 ]
-}
-
-# ── Governance gates apply (ADR-061 Rule 7) ───────────────────────────────
-
-@test "work-problems P246: pre-check governance gates apply per ADR-061 Rule 7" {
-  run grep -nE 'Governance gates apply.*ADR-061 Rule 7|graduation reinstate goes through the standard ADR-014 commit flow' "$SKILL_MD"
-  [ "$status" -eq 0 ]
-}
-
-# ── Audit trail (ADR-061 Rule 6) ──────────────────────────────────────────
-
-@test "work-problems P246: pre-check audit trail cites ADR-061 Rule 6" {
-  run grep -nE 'Audit trail.*ADR-061 Rule 6' "$SKILL_MD"
-  [ "$status" -eq 0 ]
-}
-
-@test "work-problems P246: audit trail appends README 'Recently reinstated' entry" {
-  run grep -nE 'Recently reinstated.*resolved problem-ticket ID.*Priority value|graduation criterion met.*status=resolved' "$SKILL_MD"
-  [ "$status" -eq 0 ]
-}
-
-# ── Idempotency ───────────────────────────────────────────────────────────
-
-@test "work-problems P246: pre-check is idempotent when holding-area is empty" {
-  run grep -nE 'Idempotency.*holding-area is empty|safe to invoke when holding-area is empty' "$SKILL_MD"
-  [ "$status" -eq 0 ]
-}
-
-@test "work-problems P246: pre-check is idempotent when no candidates resolve (all vp-blocked)" {
-  run grep -nE 'Safe when no candidates resolve.*all .vp-blocked.' "$SKILL_MD"
-  [ "$status" -eq 0 ]
-}
-
-# ── Holding README Process amendment ──────────────────────────────────────
-
-@test "work-problems P246: holding README Process step 5 cites P246 refined framing" {
-  run grep -nE 'refined by P246|P246 refined framing' "$HOLDING_README"
-  [ "$status" -eq 0 ]
-}
-
-@test "work-problems P246: holding README Process step 5 states criterion is positive evidence (NOT calendar)" {
-  run grep -nE 'positive evidence that the surface works as desired' "$HOLDING_README"
-  [ "$status" -eq 0 ]
-}
-
-@test "work-problems P246: holding README Process step 5 rejects calendar predicates as primary trigger" {
-  run grep -nE 'Calendar predicates.*never.*primary graduation trigger|NEVER a primary graduation trigger' "$HOLDING_README"
-  [ "$status" -eq 0 ]
-}
-
-@test "work-problems P246: holding README Process step 5 cites user direction verbatim" {
-  run grep -nE "Dogfooding makes sense, but it shouldn.t be time based" "$HOLDING_README"
-  [ "$status" -eq 0 ]
-}
-
-@test "work-problems P246: holding README Process step 5 enumerates the 3-status taxonomy" {
-  run grep -nE 'status=<resolved\|vp-blocked\|halt-no-resolution>' "$HOLDING_README"
-  [ "$status" -eq 0 ]
-}
-
-@test "work-problems P246: holding README Process step 5 explicitly preserves at-hold-time historical contracts" {
-  # Architect verdict: per-entry Currently held lines authored BEFORE
-  # the P246 framing are retained as at-hold-time historical contracts
-  # (not retroactively rewritten).
-  run grep -nE 'retained as at-hold-time historical contracts.*not retroactively rewritten' "$HOLDING_README"
-  [ "$status" -eq 0 ]
-}
-
-# ── P246 self-identification (ticket-trace) ───────────────────────────────
-
-@test "work-problems P246: SKILL.md self-identifies the cohort-graduation pre-check as P246's amendment" {
-  run grep -nE 'P246' "$SKILL_MD"
-  [ "$status" -eq 0 ]
-}
-
-@test "work-problems P246: holding README self-identifies the Process amendment as P246's refined framing" {
-  run grep -nE 'P246' "$HOLDING_README"
-  [ "$status" -eq 0 ]
-}
-
-# ── P308 self-identification (ticket-trace) ───────────────────────────────
-
-@test "work-problems P308: SKILL.md self-identifies the Rule 4 evidence-floor amendment as P308" {
-  run grep -nE 'P308' "$SKILL_MD"
-  [ "$status" -eq 0 ]
-}