@windyroad/itil 0.47.9-preview.556 → 0.47.9-preview.558

package/bin/wr-itil-check-fail-soft-skip-discipline

@@ -0,0 +1,51 @@
+#!/usr/bin/env bash
+# Generated by scripts/sync-shim-wrappers.sh from
+# packages/shared/lib/shim-wrapper-template.sh. DO NOT EDIT individual
+# shim files in packages/*/bin/wr-* directly; edit the template + run
+# `npm run sync:shim-wrappers` to regenerate.
+#
+# Resolution (ADR-080):
+#   1. If the wrapper's parent dir is semver-shaped, treat as installed-
+#      cache execution and resolve to the highest-version sibling's
+#      scripts/ entry below.
+#   2. Otherwise (parent dir is e.g. `architect`), treat as source-
+#      monorepo execution and dispatch to own scripts/. The source-repo-
+#      guard `exec` is the anchor parsed by
+#      packages/retrospective/scripts/check-tarball-shipped-shims.sh.
+#   3. If the cache parent contains zero semver-shaped siblings, exit
+#      127 with a stderr message naming the cache parent (per SQ-080-2).
+#
+# @adr ADR-080 (highest-version-wins shim wrapper plugin scaffold)
+# @adr ADR-049 (plugin-bundled scripts resolve via bin/ on $PATH — amended)
+# @problem P343 (mid-session staleness window)
+
+set -euo pipefail
+
+SHIM_DIR="$(cd "$(dirname "$0")" && pwd)"
+OWN_VERSION_DIR="$(dirname "$SHIM_DIR")"
+OWN_VERSION_NAME="$(basename "$OWN_VERSION_DIR")"
+CACHE_PARENT="$(dirname "$OWN_VERSION_DIR")"
+
+SEMVER_RE='^[0-9]+\.[0-9]+\.[0-9]+([-+][0-9A-Za-z.-]+)?$'
+
+# Source-repo guard: own parent dir is NOT semver → dispatch to own scripts/.
+if ! [[ "$OWN_VERSION_NAME" =~ $SEMVER_RE ]]; then
+  exec "$SHIM_DIR/../scripts/check-fail-soft-skip-discipline.sh" "$@"
+fi
+
+# Cache execution: pick the highest-semver sibling under CACHE_PARENT.
+HIGHEST=""
+while IFS= read -r dir; do
+  name="$(basename "$dir")"
+  [[ "$name" =~ $SEMVER_RE ]] || continue
+  if [[ -z "$HIGHEST" ]] || [[ "$(printf '%s\n%s\n' "$HIGHEST" "$name" | sort -V | tail -1)" == "$name" ]]; then
+    HIGHEST="$name"
+  fi
+done < <(find "$CACHE_PARENT" -mindepth 1 -maxdepth 1 -type d 2>/dev/null)
+
+if [[ -z "$HIGHEST" ]]; then
+  printf 'wr-shim: no cached versions in %s\n' "$CACHE_PARENT" >&2
+  exit 127
+fi
+
+exec "$CACHE_PARENT/$HIGHEST/scripts/check-fail-soft-skip-discipline.sh" "$@"

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@windyroad/itil",
-  "version": "0.47.9-preview.556",
+  "version": "0.47.9-preview.558",
   "description": "ITIL-aligned IT service management for Claude Code (problem, and future incident/change skills)",
   "bin": {
     "windyroad-itil": "./bin/install.mjs"

package/scripts/check-fail-soft-skip-discipline.sh

@@ -0,0 +1,141 @@
+#!/usr/bin/env bash
+# check-fail-soft-skip-discipline.sh — advisory lint for P351 (skills
+# fail-soft-skip when their precondition config is missing — should
+# auto-bootstrap with user input as needed rather than silently
+# skipping).
+#
+# Detects SKILL.md prose that names the fail-soft-skip pattern without
+# a paired auto-bootstrap routine. The default behaviour of skills with
+# config-file preconditions has been "fail-soft skip" when the config is
+# missing; the right behaviour (per P351) is to recognise the gap and
+# auto-bootstrap the missing config — invoking AskUserQuestion for any
+# required user input, then proceeding with the original pass. In AFK
+# mode where AskUserQuestion is unavailable, queue a config-direction
+# outstanding_question and continue other work rather than silently
+# skipping a desired capability.
+#
+# This lint walks `packages/*/skills/*/SKILL.md` files for matching
+# patterns and emits one WARN line per match. Per architect review
+# (a6747bd57c7953b14): the broad `skipping` pattern false-positives on
+# legitimate per-channel skip prose; we tighten to the
+# `skipping.*config|skipping.*not configured` shape that is the
+# load-bearing signal class.
+#
+# Usage:
+#   check-fail-soft-skip-discipline.sh [<repo-root>]
+#     <repo-root> defaults to the current working directory.
+#
+# Environment:
+#   WR_FAIL_SOFT_SKIP_WARN_ONLY=1   Phase 1 advisory (default) — exit 0
+#                                    even when violations exist.
+#   WR_FAIL_SOFT_SKIP_WARN_ONLY=0   Phase 2 load-bearing — exit 1 when
+#                                    violations exist.
+#
+# Exit codes:
+#   0 = clean OR Phase 1 advisory with violations
+#   1 = Phase 2 load-bearing with violations
+#   2 = usage / path error
+#
+# Output format (one line per violation, to stderr):
+#   WARN  <relpath>:<line>  <matched-pattern>: <line-snippet>
+#
+# Promotion criteria (Phase 1 → Phase 2):
+#   Promote `WR_FAIL_SOFT_SKIP_WARN_ONLY=0` once every affected SKILL.md
+#   has been migrated to the auto-bootstrap pattern.
+#
+# @adr ADR-040 (advisory-then-load-bearing reusable pattern)
+# @adr ADR-049 (plugin-bundled scripts; PATH shim)
+# @adr ADR-052 (behavioural-tests-default)
+# @adr ADR-013 Rule 6 (non-interactive fail-safe — advisory exit 0)
+# @jtbd JTBD-001 (Enforce Governance Without Slowing Down)
+# @jtbd JTBD-101 (Extend the Suite with New Plugins)
+# @problem P351 (skills fail-soft-skip when precondition config missing
+#                — should auto-bootstrap with user input)
+
+set -uo pipefail
+
+# Self-application: this lint grep / sed / awks SKILL.md content.
+export LC_ALL=en_US.UTF-8
+
+REPO_ROOT="${1:-$(pwd)}"
+WARN_ONLY="${WR_FAIL_SOFT_SKIP_WARN_ONLY:-1}"
+
+if [ ! -d "$REPO_ROOT" ]; then
+  echo "check-fail-soft-skip-discipline: not a directory: $REPO_ROOT" >&2
+  exit 2
+fi
+
+if [ ! -d "$REPO_ROOT/packages" ]; then
+  echo "check-fail-soft-skip-discipline: no packages/ subdir under $REPO_ROOT" >&2
+  exit 2
+fi
+
+# Patterns tightened per architect review to load-bearing signal class
+# (avoid false-positives on legitimate per-channel skip prose like
+# "skipping the failing channel/report"):
+#   1. `fail-soft skip`            — the canonical name of the pattern.
+#   2. `silently skip`             — direct synonym.
+#   3. `skipping.*config`          — skipping tied to a config artefact.
+#   4. `skipping.*not configured`  — explicit precondition-config skip.
+#   5. `not configured.*skip`      — the reverse phrasing.
+PATTERNS=(
+  'fail-soft skip'
+  'silently skip'
+  'skipping.*config'
+  'skipping.*not configured'
+  'not configured.*skip'
+)
+
+# Combined extended-regex for a single grep pass.
+PATTERN_RE='fail-soft skip|silently skip|skipping.*config|skipping.*not configured|not configured.*skip'
+
+mapfile -t TARGETS < <(
+  find "$REPO_ROOT/packages" \
+    -mindepth 4 -maxdepth 5 \
+    -type f -name 'SKILL.md' \
+    -path '*/skills/*' \
+    2>/dev/null | sort
+)
+
+violations=0
+scanned=0
+
+for file in "${TARGETS[@]}"; do
+  scanned=$((scanned + 1))
+  rel="${file#"$REPO_ROOT"/}"
+
+  # grep -n -E emits "<line_no>:<line>" per match. Read each match and
+  # emit a WARN line. We deliberately do NOT attempt to detect a paired
+  # auto-bootstrap routine in the same file — Phase 1 emits one WARN per
+  # raw pattern hit; SKILL.md authors disambiguate. The promotion
+  # criteria for Phase 2 is "every WARN'd file migrated to the
+  # auto-bootstrap pattern".
+  while IFS= read -r match; do
+    [ -z "$match" ] && continue
+    line_no="${match%%:*}"
+    line_text="${match#*:}"
+    # Derive which pattern hit. First-match-wins ordering.
+    matched_pattern=''
+    for pat in "${PATTERNS[@]}"; do
+      if echo "$line_text" | grep -E -i -q "$pat"; then
+        matched_pattern="$pat"
+        break
+      fi
+    done
+    # Trim the snippet to keep WARN lines readable.
+    snippet="${line_text:0:120}"
+    echo "WARN  $rel:$line_no  $matched_pattern: $snippet" >&2
+    violations=$((violations + 1))
+  done < <(grep -E -i -n "$PATTERN_RE" "$file" 2>/dev/null || true)
+done
+
+if [ "$violations" -gt 0 ]; then
+  echo "" >&2
+  echo "check-fail-soft-skip-discipline: $violations potential fail-soft-skip site(s) across $scanned SKILL.md file(s)" >&2
+  echo "Phase 1 advisory (WR_FAIL_SOFT_SKIP_WARN_ONLY=$WARN_ONLY). Authors should pair each site with an auto-bootstrap routine per P351." >&2
+  if [ "$WARN_ONLY" = "0" ]; then
+    exit 1
+  fi
+fi
+
+exit 0

package/scripts/test/check-fail-soft-skip-discipline.bats

@@ -0,0 +1,172 @@
+#!/usr/bin/env bats
+
+# @problem P351 — skills fail-soft-skip when their precondition config is
+#                 missing — should auto-bootstrap with user input as needed
+#                 rather than silently skipping.
+#
+# Contract: `check-fail-soft-skip-discipline.sh <repo-root>` walks
+# `<repo-root>/packages/*/skills/*/SKILL.md`, greps each file for the
+# tightened fail-soft-skip pattern set
+# (fail-soft skip|silently skip|skipping.*config|skipping.*not configured|
+#  not configured.*skip), emits one `WARN  <relpath>:<line>  <pat>: ...`
+# line per match on stderr, and exits 0 in Phase 1 advisory mode
+# regardless of whether violations were found.
+#
+# @adr ADR-049 (bin/ on PATH shim — adopter-safe script resolution)
+# @adr ADR-052 (Behavioural bats default)
+# @adr ADR-040 (Advisory-then-load-bearing reusable pattern)
+# @adr ADR-013 Rule 6 (Non-interactive fail-safe — advisory exit 0)
+# @jtbd JTBD-001 (Enforce Governance — pattern-discipline lint)
+# @jtbd JTBD-101 (Extend the Suite — extensible pattern per skill site)
+
+setup() {
+  SCRIPTS_DIR="$(cd "$(dirname "$BATS_TEST_FILENAME")/.." && pwd)"
+  SCRIPT="$SCRIPTS_DIR/check-fail-soft-skip-discipline.sh"
+  FIXTURE_ROOT="$(mktemp -d)"
+}
+
+teardown() {
+  rm -rf "$FIXTURE_ROOT"
+}
+
+# ── Existence + executable ──────────────────────────────────────────────────
+
+@test "check-fail-soft-skip-discipline: script exists" {
+  [ -f "$SCRIPT" ]
+}
+
+@test "check-fail-soft-skip-discipline: script is executable" {
+  [ -x "$SCRIPT" ]
+}
+
+# ── Phase 1 advisory exit-code contract ─────────────────────────────────────
+
+@test "Phase 1 advisory: exits 0 on a clean fixture" {
+  mkdir -p "$FIXTURE_ROOT/packages/sample/skills/clean-skill"
+  cat > "$FIXTURE_ROOT/packages/sample/skills/clean-skill/SKILL.md" <<'EOF'
+# Sample clean skill
+
+This skill auto-bootstraps any missing precondition config via
+AskUserQuestion. No fail-soft pattern present.
+EOF
+  run "$SCRIPT" "$FIXTURE_ROOT"
+  [ "$status" -eq 0 ]
+}
+
+@test "Phase 1 advisory: exits 0 even when violations present (default WR_FAIL_SOFT_SKIP_WARN_ONLY=1)" {
+  mkdir -p "$FIXTURE_ROOT/packages/sample/skills/dirty-skill"
+  cat > "$FIXTURE_ROOT/packages/sample/skills/dirty-skill/SKILL.md" <<'EOF'
+# Sample dirty skill
+
+This skill performs a fail-soft skip when the config is missing.
+EOF
+  run "$SCRIPT" "$FIXTURE_ROOT"
+  [ "$status" -eq 0 ]
+}
+
+# ── WARN-on-fixture: behavioural ────────────────────────────────────────────
+
+@test "WARN-on-fixture: matches 'fail-soft skip' literal" {
+  mkdir -p "$FIXTURE_ROOT/packages/sample/skills/dirty-skill"
+  cat > "$FIXTURE_ROOT/packages/sample/skills/dirty-skill/SKILL.md" <<'EOF'
+# Dirty skill
+
+When the file is missing emit a fail-soft skip and continue.
+EOF
+  run "$SCRIPT" "$FIXTURE_ROOT"
+  [ "$status" -eq 0 ]
+  [[ "$output" == *"WARN"* ]]
+  [[ "$output" == *"packages/sample/skills/dirty-skill/SKILL.md"* ]]
+  [[ "$output" == *"fail-soft skip"* ]]
+}
+
+@test "WARN-on-fixture: matches 'silently skip' synonym" {
+  mkdir -p "$FIXTURE_ROOT/packages/sample/skills/silent-skill"
+  cat > "$FIXTURE_ROOT/packages/sample/skills/silent-skill/SKILL.md" <<'EOF'
+# Silent skill
+
+If config missing, silently skip the pass.
+EOF
+  run "$SCRIPT" "$FIXTURE_ROOT"
+  [ "$status" -eq 0 ]
+  [[ "$output" == *"silently skip"* ]]
+}
+
+@test "WARN-on-fixture: matches tightened 'skipping.*config' shape" {
+  mkdir -p "$FIXTURE_ROOT/packages/sample/skills/skipper-skill"
+  cat > "$FIXTURE_ROOT/packages/sample/skills/skipper-skill/SKILL.md" <<'EOF'
+# Skipper skill
+
+Skipping the pass because the channels config is absent.
+EOF
+  run "$SCRIPT" "$FIXTURE_ROOT"
+  [ "$status" -eq 0 ]
+  [[ "$output" == *"skipper-skill/SKILL.md"* ]]
+}
+
+# ── CLEAN-on-fixture: behavioural negative ──────────────────────────────────
+
+@test "CLEAN-on-fixture: no WARN emitted when SKILL.md is fail-soft-skip free" {
+  mkdir -p "$FIXTURE_ROOT/packages/sample/skills/auto-bootstrap-skill"
+  cat > "$FIXTURE_ROOT/packages/sample/skills/auto-bootstrap-skill/SKILL.md" <<'EOF'
+# Auto-bootstrap skill
+
+When the config file is missing, auto-bootstrap it via AskUserQuestion
+(interactive mode) or queue a config-direction outstanding_question
+(AFK mode). The skill resumes the original pass after bootstrap.
+EOF
+  run "$SCRIPT" "$FIXTURE_ROOT"
+  [ "$status" -eq 0 ]
+  ! [[ "$output" == *"WARN"* ]]
+}
+
+@test "CLEAN-on-fixture: does NOT false-positive on legitimate per-channel skip prose (architect tightening)" {
+  mkdir -p "$FIXTURE_ROOT/packages/sample/skills/per-channel-skill"
+  cat > "$FIXTURE_ROOT/packages/sample/skills/per-channel-skill/SKILL.md" <<'EOF'
+# Per-channel skill
+
+On rate-limit, log advisory and skip that channel only — skipping the
+failing channel/report is fine because other channels proceed. This
+prose is the inverse of fail-soft-skip-on-missing-config and must NOT
+fire a WARN per the architect-tightened pattern set.
+EOF
+  run "$SCRIPT" "$FIXTURE_ROOT"
+  [ "$status" -eq 0 ]
+  # The "fail-soft-skip-on-missing-config" hyphenated literal in the
+  # prose does NOT contain a literal space-separated "fail-soft skip"
+  # phrase, but the descriptive text does. Verify the script catches the
+  # descriptive use and does NOT mistake the per-channel skip-prose for
+  # the load-bearing pattern.
+  # We expect EITHER zero WARNs (if the descriptive use is the only
+  # hit) OR exactly the descriptive lines. Either way the per-channel
+  # "skipping the failing channel/report" prose alone must NOT generate
+  # a WARN.
+  channel_only_warns=$(echo "$output" | grep -c "skipping the failing channel" || true)
+  [ "$channel_only_warns" = "0" ]
+}
+
+# ── Phase 2 promotion: behavioural ──────────────────────────────────────────
+
+@test "Phase 2 promotion: exits 1 when WR_FAIL_SOFT_SKIP_WARN_ONLY=0 + violations present" {
+  mkdir -p "$FIXTURE_ROOT/packages/sample/skills/dirty-skill"
+  cat > "$FIXTURE_ROOT/packages/sample/skills/dirty-skill/SKILL.md" <<'EOF'
+# Dirty skill
+
+Fail-soft skip when missing.
+EOF
+  WR_FAIL_SOFT_SKIP_WARN_ONLY=0 run "$SCRIPT" "$FIXTURE_ROOT"
+  [ "$status" -eq 1 ]
+}
+
+# ── Usage / path-error contract ─────────────────────────────────────────────
+
+@test "Usage: exits 2 on missing repo-root directory" {
+  run "$SCRIPT" "$FIXTURE_ROOT/does-not-exist"
+  [ "$status" -eq 2 ]
+}
+
+@test "Usage: exits 2 on repo-root without packages/ subdir" {
+  mkdir -p "$FIXTURE_ROOT/no-packages-here"
+  run "$SCRIPT" "$FIXTURE_ROOT/no-packages-here"
+  [ "$status" -eq 2 ]
+}

package/skills/manage-problem/SKILL.md

@@ -666,7 +666,12 @@ Detection is intentionally **strict** (explicit label or scoped-npm package only
   2. `Defer and note in ticket` — append a pending-upstream-report line to the ticket's `## Related` section using the stable marker `- **Upstream report pending** — external dependency identified; invoke /wr-itil:report-upstream when ready`. The marker wording is fixed so subsequent runs (and the work-problems `upstream-blocked` skip path) can detect "already noted" without re-firing.
   3. `Not actually upstream` — proceed without invocation; append the same marker with text `- **Upstream report pending** — false positive; detection misfire` so the prompt does not re-fire on later reviews.
 
-**Non-interactive (AFK) branch** (per ADR-013 Rule 6): when `AskUserQuestion` is unavailable, default to option 2 — append the pending-upstream-report line with the stable `- **Upstream report pending** —` marker. Do NOT auto-invoke `/wr-itil:report-upstream`; its Step 6 security-path branch is interactive and would halt the orchestrator anyway (per ADR-024 Consequences). The appended line lets the user see the deferred action when they return.
+**Non-interactive (AFK) branch** (per ADR-013 Rule 6 + ADR-024 2026-06-04 (P270) amendment): when `AskUserQuestion` is unavailable, **auto-invoke `/wr-itil:report-upstream`** instead of deferring with the marker. The skill composes the report draft via its own Steps 1–5/4b/5c/6 then scores the drafted prose via the `wr-risk-scorer:external-comms` agent (ADR-028) per the ADR-024 2026-06-04 amendment's orchestrator-side pre-fire gate. Branches:
+
+- **Below external-comms appetite** → the skill proceeds (public-issue path Step 5, comment path Step 5c, or security path Step 6 per the existing classification routing); commits the `## Reported Upstream` back-write per Step 7 / Step 8.
+- **Above appetite** → the skill takes risk-reducing measures (per ADR-042 within-axis precedent generalised to the external-comms risk class — the measures vocabulary is **open-ended LLM judgement** per ADR-024 2026-06-04 second-amendment ratification leaf (a): the `wr-risk-scorer:external-comms` agent's own scoring picks the remedy case-by-case, matching ADR-042's open-vocabulary precedent — NOT a bounded enumeration); re-scores; if within appetite → sends; else → **queues** an `outstanding_questions` entry naming the local ticket ID + queued report path + risk-reduce attempts + residual band + remedy ("review the queued report at `/wr-itil:report-upstream <NNN> <upstream-repo-url>` on return"). The orchestrator continues (P352 queue-and-continue). The `## Queued Upstream Report` section (renamed from `## Drafted Upstream Report` per ADR-024 2026-06-04 second-amendment leaf (c) — same shape; new name reflects the queue-for-review-on-return semantics) carries the report content for the queued question's reference. Security-path routing follows leaf (b) ratification: upstream-with-`SECURITY.md` + below-appetite → file via the declared channel; upstream-without-`SECURITY.md` but with another disclosure channel → external-comms-gated assessment considering impact to (i) our repository, (ii) our reputation, (iii) the party we are reporting to.
+
+The legacy `- **Upstream report pending** —` marker append (the pre-2026-06-04 AFK default) is **superseded** by this auto-invoke branch for all classifications including security. Tickets that already carry the marker from prior sessions are still handled correctly by the work-problems Step 4 classifier — the new path's "already-noted check" matches the legacy marker shape and routes to the report-upstream invocation. The marker shape is retained for backward compatibility on the parking + interactive fallback paths (interactive option 2 still appends it; see options 1/2/3 above).
 
 **Scope**: this detection block fires at two points —
 

package/skills/report-upstream/SKILL.md

@@ -149,7 +149,7 @@ fi
   2. `Comment on the existing upstream report` — route to Step 5c with the existing URL's issue number; appropriate when new evidence has emerged since the previous report.
   3. `File a new upstream issue anyway (override)` — explicit override after user has reviewed the existing record and judged the second filing warranted (e.g. previous report was closed without resolution and a fresh tracker is needed).
 
-**AFK / non-interactive branch** — apply the **interim static heuristic** (no subagent dispatch; the maintainer-annoyance risk evaluator is deferred per ADR-028 line 117 — see "AFK static heuristic" below). Default action: halt and save the drafted report to the local ticket's `## Drafted Upstream Report` section; do NOT auto-comment. The static heuristic remains in place until `wr-risk-scorer:external-comms` ships, at which point the AFK branch wires the gate combination (maintainer-annoyance + leak gate, both within appetite) per the ticket Direction decision (2026-04-21).
+**AFK / non-interactive branch** — apply the **interim static heuristic** (no subagent dispatch; the maintainer-annoyance risk evaluator is deferred per ADR-028 line 117 — see "AFK static heuristic" below). Default action: halt and save the drafted report to the local ticket's `## Queued Upstream Report` section; do NOT auto-comment. The static heuristic remains in place until `wr-risk-scorer:external-comms` ships, at which point the AFK branch wires the gate combination (maintainer-annoyance + leak gate, both within appetite) per the ticket Direction decision (2026-04-21).
 
 #### 4b.2. Third-party search
 
@@ -197,20 +197,20 @@ If Stage 2 produces one or more `same-problem` matches, surface them to the user
 
 `uncertain` matches surface alongside `same-problem` matches with their verdict labelled, so the user can review. The skill never auto-resolves an `uncertain` verdict.
 
-**AFK / non-interactive branch** — apply the same interim static heuristic as 4b.1: halt and save the drafted report to the local ticket's `## Drafted Upstream Report` section. The third-party-match auto-comment path requires the deferred `wr-risk-scorer:external-comms` gate (maintainer-annoyance + leak), so the AFK branch must NOT auto-comment under the static heuristic.
+**AFK / non-interactive branch** — apply the same interim static heuristic as 4b.1: halt and save the drafted report to the local ticket's `## Queued Upstream Report` section. The third-party-match auto-comment path requires the deferred `wr-risk-scorer:external-comms` gate (maintainer-annoyance + leak), so the AFK branch must NOT auto-comment under the static heuristic.
 
 #### AFK static heuristic (interim, until `wr-risk-scorer:external-comms` ships)
 
 The Direction decision (2026-04-21) pins the AFK auto-comment branch on **two gates passing together**: the maintainer-annoyance risk evaluator AND the P064 external-comms leak gate, both within RISK-POLICY.md's commit-layer appetite (Low, ≤4/25). Neither gate exists yet — ADR-028 declares the `wr-risk-scorer:external-comms` subagent type but P064's implementation is open at WSJF 3.0 (Effort L), and the maintainer-annoyance evaluator was deferred by architect review on P070 to compose with the same subagent rather than ship as a separate evaluator (per ADR-028 line 117 — *"Third evaluator (licence-compliance, etc.) adding to the same gate — when it emerges, amend this ADR's evaluator list and the composite marker's `evaluator_set` component; no new ADR expected."*).
 
-**Static heuristic, valid until both gates ship**: in AFK mode, both 4b.1 and 4b.2 default to **halt and save the drafted report**. No auto-comment, no auto-file. The drafted report is appended to the local ticket's `## Drafted Upstream Report` section so the user can review and act manually on return. This matches JTBD-006's "does not trust the agent to make judgement calls" stance — the conservative default is the right interim behaviour.
+**Static heuristic, valid until both gates ship**: in AFK mode, both 4b.1 and 4b.2 default to **halt and save the drafted report**. No auto-comment, no auto-file. The drafted report is appended to the local ticket's `## Queued Upstream Report` section so the user can review and act manually on return. This matches JTBD-006's "does not trust the agent to make judgement calls" stance — the conservative default is the right interim behaviour.
 
 **Re-wire trigger**: when `wr-risk-scorer:external-comms` lands (ADR-028 implementation, P064 closure), amend this section to invoke both evaluators and proceed with auto-comment ONLY when both verdicts return PASS within appetite. Update the AFK behaviour summary table accordingly. Until then, the static heuristic stands.
 
-**Drafted Upstream Report save format** (used by both 4b.1 and 4b.2 AFK halts; mirrors the security-path halt pattern from Step 6 per ADR-024 Consequences lines 116, 123):
+**Queued Upstream Report save format** (used by both 4b.1 and 4b.2 AFK halts; mirrors the security-path halt pattern from Step 6 per ADR-024 Consequences lines 116, 123. **Renamed from `## Drafted Upstream Report` per ADR-024 2026-06-04 second-amendment ratification (c)** — same shape; new name reflects post-amendment semantics, queue-for-review-on-return not loop-stopping halt-and-save):
 
 ```markdown
-## Drafted Upstream Report
+## Queued Upstream Report
 
 - **Drafted**: <YYYY-MM-DD>
 - **Target upstream**: <upstream-repo-url>
@@ -460,7 +460,7 @@ Parse for a disclosure channel:
   }
   EOF
   ```
-- **`security@` mailbox** (or any `mailto:` link): **halt** and surface the mailbox + drafted report to the user. Do NOT auto-send email — out of scope, no infra. Save the drafted report to `docs/problems/<NNN>-<title>.<status>.md`'s `## Drafted Upstream Report` appendage section so the user can copy + send.
+- **`security@` mailbox** (or any `mailto:` link): **halt** and surface the mailbox + drafted report to the user. Do NOT auto-send email — out of scope, no infra. Save the drafted report to `docs/problems/<NNN>-<title>.<status>.md`'s `## Queued Upstream Report` appendage section so the user can copy + send.
 - **Other documented channel** (Tidelift, HackerOne, vendor-specific URL): halt and surface the channel + drafted report.
 
 If upstream has **NO `SECURITY.md`** (404):
@@ -468,9 +468,9 @@ If upstream has **NO `SECURITY.md`** (404):
   - `(a) Open a private GitHub Security Advisory` — uses `gh api repos/.../security-advisories` against the upstream if it's GitHub-hosted.
   - `(b) Contact the maintainer out-of-band first` — halt, no automated action.
   - `(c) Downgrade the classification (your judgement)` — re-route via the public-issue path in Step 5.
-- **AFK / non-interactive context**: do NOT auto-resolve. Save the drafted report to the local ticket's `## Drafted Upstream Report` section and **halt the orchestrator** — this is a loop-stopping event per ADR-024 Consequences. AFK orchestrators must never auto-report a security-classified ticket.
+- **AFK / non-interactive context** (per ADR-024 2026-06-04 (P270) amendment — **supersedes the pre-amendment "must never auto-report a security-classified ticket" rule**): score the drafted prose via `wr-risk-scorer:external-comms` (ADR-028). **Per ADR-024 2026-06-04 second-amendment leaf (b) — ratified 2026-06-04**: if upstream has a `SECURITY.md` declaring a channel AND scored below-appetite → file via that channel (Step 6 routing). If upstream has NO `SECURITY.md` but another disclosure channel exists → run the external-comms risk assessment considering impact to (i) our repository, (ii) our reputation, (iii) the party we are reporting to. Below-appetite → proceed (save the report to `## Queued Upstream Report` if the channel is non-automatable like a mailbox — the channel-action stays user-side per the no-infra-for-email constraint) + queue `outstanding_questions` entry naming the channel. Above-appetite → risk-reduce + re-score then queue per the same shape (the risk-reducing-measures vocabulary is **open-ended LLM judgement** per leaf (a) — the `wr-risk-scorer:external-comms` agent's own scoring picks the remedy case-by-case; not a bounded enumeration). Orchestrator continues per P352 queue-and-continue. The pre-amendment "AFK orchestrators must never auto-report" / "Never auto-open a public issue for a security-classified ticket" rules are **superseded** — both are now external-comms-gated per the 2026-06-04 amendment.
 
-**Never auto-open a public issue for a security-classified ticket.**
+**[Legacy phrase retained for traceability — superseded by ADR-024 2026-06-04 amendment]** Never auto-open a public issue for a security-classified ticket — the pre-amendment ban is now subordinated to the external-comms-gated per-classification branching above.
 
 ### 7. Cross-reference back-write
 
@@ -496,7 +496,7 @@ After the upstream issue or advisory is created (or drafted-and-saved in the sec
 
 Follow the ADR-014 ordering:
 
-1. `git add docs/problems/<NNN>-<title>.<status>.md` (and any `## Drafted Upstream Report` appendage if security-path halt fired).
+1. `git add docs/problems/<NNN>-<title>.<status>.md` (and any `## Queued Upstream Report` appendage if security-path halt fired).
 2. Score commit/push/release risk via `wr-risk-scorer:pipeline` subagent (or fall back to `/wr-risk-scorer:assess-release` skill per ADR-015).
 3. `git commit -m "docs(problems): P<NNN> reported upstream — <one-line summary>"`.
 
@@ -504,14 +504,14 @@ If the cumulative pipeline risk lands above appetite and `AskUserQuestion` is un
 
 ## AFK behaviour summary
 
-Five distinct AFK branches per the architect reviews of ADR-024, ADR-013 Rule 6, and the P070 dedup amendment:
+Five distinct AFK branches; **per the ADR-024 2026-06-04 (P270) amendment, ALL pre-commit branches now route through the `wr-risk-scorer:external-comms` gate** (ADR-028) — below-appetite proceeds, above-appetite risk-reduces then queues. The legacy "halt the orchestrator" semantics for dedup-match and security-path-without-declared-channel are **superseded** by queue-and-continue per P352:
 
 | Branch | AFK behaviour | Authority |
 |---|---|---|
-| Public-issue path (Step 5) | Proceeds. Voice-tone gate per ADR-028 may delegate-and-retry; that is the expected extra turn. | ADR-028 line 126 |
-| Dedup match — Step 4b halt (own re-run OR third-party `same-problem`) | Save drafted report to local ticket's `## Drafted Upstream Report` section. **Halt the orchestrator** — loop-stopping event. Interim static heuristic; auto-comment branch deferred until `wr-risk-scorer:external-comms` ships (ADR-028 line 117). | ADR-024 amendment 2026-04-25 (P070); Direction decision 2026-04-21 |
-| Security path with declared channel (Step 6, GitHub Advisories) | Proceeds via `gh api .../security-advisories`. | ADR-024 Decision Outcome step 6 |
-| Security path with `security@` / other / missing-SECURITY.md (Step 6) | Save drafted report to local ticket's `## Drafted Upstream Report` section. **Halt the orchestrator** — loop-stopping event. AFK orchestrators must never auto-report a security-classified ticket. | ADR-024 Consequences lines 116, 123 |
+| Public-issue path (Step 5) | Score drafted prose via `wr-risk-scorer:external-comms` (ADR-028). Below-appetite → proceed via `gh issue create`. Above-appetite → risk-reduce + re-score; if within → proceed; else → save draft to `## Queued Upstream Report` + queue `outstanding_questions` entry; orchestrator continues. Risk-reducing measures vocabulary is **open-ended LLM judgement** per ADR-024 2026-06-04 second-amendment leaf (a) — `wr-risk-scorer:external-comms` picks the remedy case-by-case. Voice-tone gate per ADR-028 may also delegate-and-retry on the proceed path. | ADR-024 2026-06-04 amendment (P270); ADR-024 2026-06-04 second-amendment (leaf a — open vocabulary); ADR-028 line 126 |
+| Dedup match — Step 4b (own re-run OR third-party `same-problem`) | Score the proposed comment body via `wr-risk-scorer:external-comms`. Below-appetite → proceed via `gh issue comment` (Step 5c). Above-appetite → risk-reduce + re-score (open-ended LLM judgement per leaf (a)); if within → proceed; else → save draft to `## Queued Upstream Report` + queue `outstanding_questions` entry; orchestrator continues. **The 2026-04-25 (P070) "interim static heuristic in force until that subagent ships" deferral is LIFTED** — the subagent (`packages/risk-scorer/agents/external-comms.md`) has shipped. | ADR-024 2026-06-04 amendment (P270); ADR-024 2026-04-25 amendment (P070); ADR-024 2026-06-04 second-amendment (leaf a) |
+| Security path with declared channel (Step 6, GitHub Advisories — upstream has `SECURITY.md`) | Per ADR-024 2026-06-04 second-amendment leaf (b) — ratified: if upstream has `SECURITY.md` AND below-appetite → **file** via the SECURITY.md-declared channel. Above-appetite → risk-reduce + re-score (open-ended LLM judgement per leaf (a)); if within → proceed; else → save draft + queue; orchestrator continues. | ADR-024 2026-06-04 amendment (P270); ADR-024 2026-06-04 second-amendment (leaf b); ADR-024 Decision Outcome step 6 |
+| Security path with `security@` / other / missing-SECURITY.md (Step 6) | Per ADR-024 2026-06-04 second-amendment leaf (b) — ratified: when upstream has NO `SECURITY.md` but another disclosure channel exists, score drafted prose via `wr-risk-scorer:external-comms` considering impact to (i) our repository, (ii) our reputation, (iii) the party we are reporting to. Below-appetite → save drafted report to `## Queued Upstream Report` + queue `outstanding_questions` entry naming the channel the user must follow on return (the no-infra-for-email constraint still holds — the channel-action remains user-side, but the queue surface replaces the loop-stopping halt). Above-appetite → risk-reduce + re-score (open-ended LLM judgement per leaf (a)) then queue per the same shape. Orchestrator continues. The pre-2026-06-04 "AFK orchestrators must never auto-report a security-classified ticket" rule is **superseded** by the external-comms-gated per-classification branching. | ADR-024 2026-06-04 amendment (P270); ADR-024 2026-06-04 second-amendment (leaf b); ADR-024 Consequences lines 116, 123 (superseded) |
 | Above-appetite commit (Step 8) | Skip the commit, report uncommitted state. | ADR-013 Rule 6 |
 
 ## References

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

@@ -82,6 +82,56 @@ setup() {
   [ "$status" -eq 0 ]
 }
 
+# ─── ADR-024 2026-06-04 (P270) amendment: external-comms-gated AFK auto-fire ──
+#
+# The 2026-06-04 amendment reverses the pre-amendment blanket-defer in AFK
+# (line 137 "AFK orchestrators should never auto-report a security-classified
+# ticket") to a per-classification external-comms gated branching. Below-
+# appetite proceeds; above-appetite risk-reduces then queues per P352.
+
+@test "report-upstream: SKILL.md AFK behaviour summary cites external-comms gate as authority (ADR-024 2026-06-04 P270 amendment)" {
+  # The external-comms gate must appear in the AFK summary table as the
+  # per-classification routing authority — at minimum on the public-issue,
+  # dedup, security-with-channel, and security-without-channel rows.
+  run grep -iE 'wr-risk-scorer:external-comms' "$SKILL_MD"
+  [ "$status" -eq 0 ]
+  # At least 4 occurrences in SKILL.md (one per AFK row in the summary plus
+  # at least one prose occurrence in Step 6 prose).
+  hits=$(grep -ic 'wr-risk-scorer:external-comms' "$SKILL_MD")
+  [ "$hits" -ge 4 ]
+}
+
+@test "report-upstream: SKILL.md AFK summary cites P270 amendment as authority (ADR-024 2026-06-04)" {
+  run grep -iE '2026-06-04.*P270|P270.*2026-06-04' "$SKILL_MD"
+  [ "$status" -eq 0 ]
+}
+
+@test "report-upstream: SKILL.md AFK above-appetite branch queues per P352 queue-and-continue (P270 amendment)" {
+  # The above-appetite branch on ANY AFK row must reference queue-and-continue
+  # (P352) rather than halt-the-orchestrator semantics for the pre-amendment
+  # blanket-defer rule.
+  run grep -iE 'P352|queue-and-continue' "$SKILL_MD"
+  [ "$status" -eq 0 ]
+}
+
+@test "report-upstream: SKILL.md marks the pre-amendment never-auto-public ban as superseded (P270 amendment)" {
+  # The legacy "Never auto-open a public issue for a security-classified
+  # ticket" prose is retained for traceability but must be marked as
+  # superseded by the 2026-06-04 amendment so readers do not act on the
+  # superseded rule.
+  run grep -iE 'superseded.*2026-06-04|2026-06-04.*supersed' "$SKILL_MD"
+  [ "$status" -eq 0 ]
+}
+
+@test "report-upstream: SKILL.md drops the 'interim static heuristic' deferral on the dedup branch (P070 lift-condition met per P270 amendment)" {
+  # The 2026-04-25 (P070) amendment named the dedup-AFK as an "interim static
+  # heuristic in force until wr-risk-scorer:external-comms ships". The agent
+  # has shipped; the 2026-06-04 (P270) amendment lifts the deferral. The
+  # AFK summary's dedup-path row must cite the LIFT, not the deferral.
+  run grep -iE 'interim static heuristic.*lifted|deferral.*lifted|LIFTED' "$SKILL_MD"
+  [ "$status" -eq 0 ]
+}
+
 # ─── ADR-033 problem-first classifier contract (P067) ──────────────────────────
 #
 # ADR-033 partially supersedes ADR-024 Decision Outcome Steps 3 + 5 with a

package/skills/review-problems/SKILL.md

@@ -125,7 +125,39 @@ Per ADR-062 (peer of ADR-024). Polls configured upstream channels, runs each unm
 
 #### 4.5a. Read channel config + parse invocation flags
 
-Read `docs/problems/.upstream-channels.json`. If missing or malformed: log an advisory note (`channel config absent or malformed; inbound-discovery skipped this pass`) and skip Step 4.5 entirely. Adopters who don't ship this file inherit zero ceremony tax — the downstream-adopter non-obligation per ADR-062 § Downstream-adopter contract + JTBD-101.
+Read `docs/problems/.upstream-channels.json`. Branch on state:
+
+- **File exists and parses cleanly** → continue to 4.5b with the parsed `channels[]` list.
+- **File exists but is malformed JSON** → log an advisory note (`channel config malformed JSON; inbound-discovery skipped this pass — fix the file then re-invoke`) and skip Step 4.5. This is genuine fail-soft because the user already SHIPPED a config and the malformation is an editing artefact best resolved by the user reading the parse error; auto-rewriting their file would destroy their work.
+- **File does NOT exist** → run the **auto-bootstrap routine** below per P351 / JTBD-101 / JTBD-007. **Adopters who never want to configure inbound-discovery** can keep `.upstream-channels.json` absent by answering `decline` at the interactive prompt OR allowing the AFK-mode outstanding-question to lapse; the absence stays zero-ceremony-tax (ADR-062 § Downstream-adopter non-obligation), but the silent skip is replaced by an explicit one-time-per-session prompt so the adopter has a visible signal that the inbound-discovery capability exists. <!-- @jtbd JTBD-101 (Extend the Suite — deliver-installed-features signal) --> <!-- @jtbd JTBD-007 (Keep Plugins Current — process reports what configured) --> <!-- @problem P351 (auto-bootstrap on missing precondition config) -->
+
+**Auto-bootstrap routine (P351)**: replaces the prior "missing file → silent skip" behaviour. The routine branches on AskUserQuestion availability per ADR-013 Rule 6 + ADR-044 category 1 (direction-setting):
+
+- **Interactive mode** (AskUserQuestion available):
+  1. Fire **one** `AskUserQuestion` per skill invocation (NOT per pass — adopters who decline at run 1 are not re-prompted within the same session) with options: `Bootstrap now (recommended)` / `Decline (skip inbound-discovery this session)` / `Decline permanently (write empty channels stub)`.
+  2. On `Bootstrap now`: fire a second `AskUserQuestion` for channel-type (single-select: `github-issues` / `github-discussions` / `github-security-advisories`); then a third for the per-channel coordinates (repo `<owner>/<name>` for all three; `label` for github-issues; `category` for github-discussions). For the per-coordinate prompt fire ONE multi-part `AskUserQuestion` (multiple Question objects in a single call per ADR-013 Rule 1 batched ≤4) — do NOT serialise to N round-trips.
+  3. **Preview before write** (JTBD persona-fit constraint from review): emit the planned JSON contents to the agent's user-visible output so the adopter can read it before the write fires. Default `ttl_seconds: 86400` (24h — matches ADR-062's documented TTL). Channel schema mirrors the polled-channels list at 4.5c (`type` + `repo` + per-type identifier).
+  4. Write `docs/problems/.upstream-channels.json` with the bootstrapped channel + the defaulted TTL.
+  5. **Resume the original pass** at 4.5b with the freshly-written config.
+  - On `Decline (skip inbound-discovery this session)`: log advisory (`inbound-discovery bootstrap declined this session; will re-offer next invocation`) and skip Step 4.5.
+  - On `Decline permanently (write empty channels stub)`: write `{"channels": [], "ttl_seconds": 86400, "declined_at": "<ISO>"}` so future invocations parse cleanly + skip silently. Per ADR-062 § Downstream-adopter non-obligation — the empty-channels stub IS the documented "I never want this" surface.
+
+- **AFK mode** (AskUserQuestion unavailable, e.g. invoked from `/wr-itil:work-problems`):
+  1. Log advisory (`inbound-discovery: channel config absent; queued config-direction outstanding_question, skipping THIS pass to allow other passes to proceed`).
+  2. Queue a `direction` entry per `/wr-itil:work-problems` SKILL.md Step 5 `outstanding_questions` schema (ADR-044 category 1):
+
+     ```
+     {
+       category: "direction",
+       question: "Configure inbound-discovery channels in docs/problems/.upstream-channels.json? (channel-type: github-issues | github-discussions | github-security-advisories; per-channel: repo + label/category; ttl_seconds default 86400)",
+       context: "/wr-itil:review-problems Step 4.5a: precondition config missing — auto-bootstrap blocked by AFK AskUserQuestion unavailability per ADR-013 Rule 6; deferring to loop-end Step 2.5 batched AskUserQuestion",
+       ticket_id: "<the iter's ticket>"
+     }
+     ```
+
+  3. **Continue Step 4.5 for THIS pass** with the missing-channels skip (other passes in the review proceed normally per the Fail-soft contract at 4.5 head). Do NOT halt the iter — the loop-end Step 2.5 batched `AskUserQuestion` is the documented surfacing point per the AFK contract.
+
+The routine preserves the ADR-062 § Downstream-adopter non-obligation (adopters CAN decline) while honouring P351 (adopters can no longer be silently under-delivered). Phase 1 lint at `wr-itil-check-fail-soft-skip-discipline` flags this site's adjacent siblings for follow-on remediation.
 
 Parse `$ARGUMENTS` as a whitespace-separated token list. Recognised invocation flags for inbound-discovery:
 

package/skills/work-problems/SKILL.md

@@ -335,7 +335,7 @@ Read the problem file and apply these deterministic rules:
 | Problem previously attempted twice without progress in this session | **Skip** — mark as stuck, needs interactive attention | user-answerable (direction) |
 | Open problem with outstanding user-answerable design question (naming, direction, pacing, scope) | **Skip** — surface the question at stop (Step 2.5) | user-answerable (design) |
 | Open problem needing architect design judgment (new-ADR-level question) | **Skip** — note the architect-design blocker; Step 2.5 may elevate via a pre-triggered architect call in `--deep-stop` mode | architect-design |
-| Open problem blocked on upstream dependency or Claude Code capability gap | **Skip** — but first append the pending-upstream-report marker to the ticket's `## Related` section (see P063 — run the manage-problem SKILL.md external-root-cause detection AFK fallback before skipping). The marker wording is fixed: `- **Upstream report pending** — external dependency identified; invoke /wr-itil:report-upstream when ready`. Use the already-noted check to avoid duplicates. | upstream-blocked |
+| Open problem blocked on upstream dependency or Claude Code capability gap | **Auto-invoke `/wr-itil:report-upstream` via the AFK fallback** (per ADR-024 2026-06-04 (P270) amendment — manage-problem Step 6 external-root-cause detection AFK fallback owns the actual invocation; this row routes through it). The report-upstream skill composes the draft then scores the prose via `wr-risk-scorer:external-comms` (ADR-028); below-appetite → sends; above-appetite → risk-reduces (open-ended LLM judgement per ADR-024 2026-06-04 second-amendment leaf (a)) then re-scores → sends-or-queues. Security routing per leaf (b): upstream-with-`SECURITY.md` + below-appetite → files via declared channel; upstream-without-`SECURITY.md` → external-comms-gated impact assessment to (i) our repo, (ii) our reputation, (iii) reported party. Queued reports save to `## Queued Upstream Report` (renamed from `## Drafted Upstream Report` per leaf (c)). Queue does NOT halt — outstanding_question surfaces at Step 2.4 / Step 2.5b end-of-loop per P352. Iter still classifies the ticket as `upstream-blocked` (the local ticket itself is still blocked on the upstream fix) and **skips work on it** after the report-upstream invocation completes — the report-upstream call is the action this row takes; classification stays `upstream-blocked` so Step 4 routes to skip-rather-than-work. Tickets already carrying `- **Upstream report pending** —` from prior sessions are detected via the already-noted check and routed to the report-upstream invocation (the marker shape is retained as the detection substrate per the 2026-06-04 amendment). | upstream-blocked |
 
 The default is to work the problem. Only skip when the rule explicitly says so. This is an AFK loop — forward progress matters more than avoiding dead ends, because dead ends are cheap (findings are saved) and interactive input is expensive (user is absent).
 
@@ -343,7 +343,7 @@ The default is to work the problem. Only skip when the rule explicitly says so.
 
 - **user-answerable** — the user can answer directly (verification, naming, direction, pacing, scope). Step 2.5 surfaces these as questions (interactive) or in the Outstanding Design Questions table (non-interactive / AFK).
 - **architect-design** — requires architect judgment first; may escalate to a new ADR. Step 2.5 can optionally pre-trigger the architect agent in `--deep-stop` mode to produce a concrete user-answerable question. Otherwise noted as "pending architect review".
-- **upstream-blocked** — external dependency, Claude Code capability gap, or waiting on third-party fix. Truly terminal for this loop — no user question would change anything. Report the blocker and move on. **Before skipping, run the manage-problem external-root-cause detection AFK fallback** (per P063): grep the ticket for the stable marker `- **Upstream report pending** —` or `- **Reported Upstream:**` / a `## Reported Upstream` section; if none is present, append `- **Upstream report pending** — external dependency identified; invoke /wr-itil:report-upstream when ready` to the ticket's `## Related` section. This preserves the outbound audit trail across AFK iterations so the user can see the deferred action on return.
+- **upstream-blocked** — external dependency, Claude Code capability gap, or waiting on third-party fix. Truly terminal for this loop — no user question would change anything. Report the blocker (now via auto-invoke of `/wr-itil:report-upstream`, per ADR-024 2026-06-04 (P270) amendment) and move on. **Before skipping, run the manage-problem external-root-cause detection AFK fallback** (per P063 amended 2026-06-04): the fallback now invokes `/wr-itil:report-upstream` rather than only appending the marker. The report-upstream skill scores the drafted prose via `wr-risk-scorer:external-comms` (ADR-028); below-appetite branches send (public-issue Step 5 / comment Step 5c / security Step 6 per classification); above-appetite branches risk-reduce + re-score; if-still-above queue an `outstanding_questions` entry per P352 queue-and-continue (orchestrator does NOT halt). Existing tickets carrying `- **Upstream report pending** —` or `- **Reported Upstream:**` / a `## Reported Upstream` section are detected via the already-noted check; the marker shape is retained for backward compatibility and as the detection substrate. The outbound audit trail across AFK iterations now reflects ACTUAL filings (or queued-for-review drafts), not just deferred intents.
 
 Record the category alongside the skip reason in the iteration report so Step 2.5 can read the categories deterministically.
 
@@ -666,14 +666,24 @@ After the iteration's commit lands but before starting the next iteration, check
 **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`** — graduate. Per ADR-061 Rule 5 + ADR-013 Rule 5, this is policy-authorised silent proceed (no `AskUserQuestion`). 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>); class <3a|3b>; evidence cited`. For class=3b cohorts, ALL members of the cohort with `status=resolved` graduate together atomically (Rule 3b cohort propagation — entire cohort ships or none does). Amend the iter's main commit per ADR-042 Rule 3 amend-based folding to preserve the ADR-032 one-commit-per-iteration invariant.
+   - **`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.
+**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.
 
@@ -818,7 +828,7 @@ When `AskUserQuestion` is unavailable or the user is AFK, the skill (and the del
 | 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` | Graduate. Per ADR-061 Rule 5 + ADR-013 Rule 5 policy-authorised silent proceed (no `AskUserQuestion`): `git mv docs/changesets-holding/<basename> .changeset/<basename>`, append README "Recently reinstated" entry, amend the iter's main commit per ADR-042 Rule 3. For class=3b cohorts, all cohort members with `status=resolved` graduate atomically (Rule 3b cohort propagation). Per ADR-061 Rule 5 + Rule 6 + Rule 7 + P246 (Step 6.5 Cohort-graduation pre-check). 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 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. |
 | 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). |
@@ -833,7 +843,7 @@ 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 | Append the stable `- **Upstream report pending** — external dependency identified; invoke /wr-itil:report-upstream when ready` marker to the ticket's `## Related` section; do NOT auto-invoke `/wr-itil:report-upstream` (Step 6 security-path branch is interactive — per ADR-024 Consequences). Use the already-noted grep check to avoid duplicate lines. Per P063 + ADR-013 Rule 6. |
-| Mid-loop ask between iters in the orchestrator's main turn | Forbidden except at framework-prescribed halt 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 6.75 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 6.5 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 discipline (orchestrator main turn) — P130
 
@@ -847,6 +857,7 @@ The orchestrator MUST NOT call `AskUserQuestion` between iterations except at th
 - **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).
 

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

@@ -88,19 +88,22 @@ setup() {
 
 # ── Three-status branching contract ───────────────────────────────────────
 
-@test "work-problems P246: pre-check branches on status=resolved → graduate" {
-  # The load-bearing positive contract: status=resolved means graduate
-  # via git mv + README append + ADR-042 Rule 3 amend.
-  run grep -nE '`status=resolved`.*graduate' "$SKILL_MD"
+@test "work-problems P246/P308: status=resolved routes to Rule 4 evidence-floor judgement (NOT auto-graduate)" {
+  # P308 amendment: status=resolved is necessary-but-not-sufficient.
+  # The evaluator script disclaims Rule 4 (LLM-owned). SKILL prose must
+  # surface Rule 4 evidence-floor judgement before any git mv.
+  run grep -nE '`status=resolved`.*route to.*Rule 4 evidence-floor judgement' "$SKILL_MD"
   [ "$status" -eq 0 ]
 }
 
-@test "work-problems P246: status=resolved branch performs git mv from holding to .changeset/" {
+@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 P246: status=resolved branch amends the iter commit per ADR-042 Rule 3" {
+@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 ]
 }
@@ -194,23 +197,49 @@ setup() {
   [ "$status" -eq 0 ]
 }
 
-# ── Policy authorisation (ADR-013 Rule 5 + ADR-061 Rule 5) ────────────────
+# ── Policy authorisation (ADR-013 Rules 1/5/6 + ADR-061 Rules 4/5 + P352) ─
 
-@test "work-problems P246: resolved-branch is policy-authorised silent proceed (no AskUserQuestion)" {
-  run grep -nE 'policy-authorised silent proceed.*no.*AskUserQuestion' "$SKILL_MD"
+@test "work-problems P308: pre-check cites ADR-061 Rule 4 as the LLM-owned evidence-floor surface" {
+  # Rule 4 evidence-floor judgement is class-specific and LLM-owned per
+  # ADR-044 framework-resolution boundary; surfacing it as AskUserQuestion
+  # (interactive) or queue (AFK) is the load-bearing delegation surface.
+  run grep -nE 'ADR-061 Rule 4' "$SKILL_MD"
   [ "$status" -eq 0 ]
 }
 
-@test "work-problems P246: pre-check cites ADR-013 Rule 5" {
-  run grep -nE 'ADR-013 Rule 5' "$SKILL_MD"
+@test "work-problems P308: pre-check cites ADR-013 Rule 6 as the AFK queue-and-continue fail-safe" {
+  # Per P352 user-ratified universal default: when AskUserQuestion is
+  # unavailable (AFK iter), queue the question + continue — do NOT halt,
+  # do NOT silently fail-soft, do NOT auto-default.
+  run grep -nE 'ADR-013 Rule 6.*P352|P352.*ADR-013 Rule 6' "$SKILL_MD"
   [ "$status" -eq 0 ]
 }
 
-@test "work-problems P246: pre-check cites ADR-061 Rule 5" {
+@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" {
@@ -288,3 +317,10 @@ setup() {
   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 ]
+}