@mmerterden/multi-agent-pipeline 8.6.2 → 10.0.6

package/pipeline/scripts/smoke-phase4-triage.sh

@@ -1,15 +1,15 @@
 #!/usr/bin/env bash
-# smoke-phase4-triage.sh — contract check for Phase 4 triage output.
+# smoke-phase4-triage.sh  -  contract check for Phase 4 triage output.
 #
 # Does NOT hit real review agents. Instead:
 #   1. Synthesizes representative triage-output fixtures (valid + each edge case).
 #   2. Validates them against pipeline/schemas/triage-output.schema.json
-#      using a minimal JSON-structure check (zero-dep — same style as
+#      using a minimal JSON-structure check (zero-dep  -  same style as
 #      validate-schemas.mjs).
 #   3. Confirms the pipeline's documented edge-case rules hold for:
 #      - no-findings short-circuit (approved: true)
 #      - over-rejection guard threshold (>0.80 rejected)
-#      - schema rejects invented reviewer (e.g. "haiku" — removed in v2.1.0)
+#      - schema rejects invented reviewer (e.g. "haiku"  -  removed in v2.1.0)
 #      - accepted finding without `fix` is rejected (fix is required on accepted)
 
 set -euo pipefail
@@ -71,7 +71,7 @@ EOF
 check_via_validator "$TMP/accepted.json" 0 "accepted blocking with fix + reviewer is valid"
 
 echo ""
-echo "→ fixture 3: accepted without fix (must reject — exit 1)"
+echo "→ fixture 3: accepted without fix (must reject  -  exit 1)"
 cat > "$TMP/no-fix.json" <<'EOF'
 {
   "accepted": [
@@ -85,7 +85,7 @@ EOF
 check_via_validator "$TMP/no-fix.json" 1 "accepted without fix triggers exit 1"
 
 echo ""
-echo "→ fixture 4: reviewer=haiku (removed in v2.1.0, must reject — exit 1)"
+echo "→ fixture 4: reviewer=haiku (removed in v2.1.0, must reject  -  exit 1)"
 cat > "$TMP/haiku.json" <<'EOF'
 {
   "accepted": [
@@ -99,7 +99,7 @@ EOF
 check_via_validator "$TMP/haiku.json" 1 "haiku reviewer triggers exit 1"
 
 echo ""
-echo "→ fixture 5: over-rejection guard (5 rejected, 0 others — exit 2)"
+echo "→ fixture 5: over-rejection guard (5 rejected, 0 others  -  exit 2)"
 cat > "$TMP/over-reject.json" <<'EOF'
 {
   "accepted": [],
@@ -117,7 +117,7 @@ EOF
 check_via_validator "$TMP/over-reject.json" 2 "100% rejection triggers over-rejection guard (exit 2)"
 
 echo ""
-echo "→ fixture 6: contradiction (approved=false but no blocker — exit 3 with corrected)"
+echo "→ fixture 6: contradiction (approved=false but no blocker  -  exit 3 with corrected)"
 cat > "$TMP/contradiction.json" <<'EOF'
 {
   "accepted": [
@@ -137,6 +137,93 @@ else
   fail "corrected.approved was not true"
 fi
 
+echo ""
+echo "→ fixture 7: optional consensus block, unverified verdict (valid  -  exit 0)"
+cat > "$TMP/consensus-unverified.json" <<'EOF'
+{
+  "accepted": [],
+  "deferred": [],
+  "rejected": [],
+  "approved": true,
+  "consensus": {
+    "reviewerCount": 2,
+    "verdict": "unverified",
+    "disagreements": [
+      {"file":"Auth/KeychainStore.swift","line":40,"issue":"token persisted without access-control flag","note":"both approved a keychain change  -  agreement unverified, confirm manually"}
+    ]
+  }
+}
+EOF
+check_via_validator "$TMP/consensus-unverified.json" 0 "optional consensus (unverified) is valid"
+
+echo ""
+echo "→ fixture 8: consensus split verdict (valid  -  exit 0)"
+cat > "$TMP/consensus-split.json" <<'EOF'
+{
+  "accepted": [
+    {"severity":"blocking","file":"src/pay.ts","line":7,"issue":"amount parsed as float loses precision","fix":"use integer minor units","reviewer":"opus"}
+  ],
+  "deferred": [],
+  "rejected": [],
+  "approved": false,
+  "consensus": {
+    "reviewerCount": 2,
+    "verdict": "split",
+    "disagreements": [
+      {"file":"src/pay.ts","line":7,"issue":"float money parsing","note":"Opus blocking, Sonnet approved"}
+    ]
+  }
+}
+EOF
+check_via_validator "$TMP/consensus-split.json" 0 "consensus split verdict is valid"
+
+echo ""
+echo "→ fixture 9: bad consensus verdict enum (must reject  -  exit 1)"
+cat > "$TMP/consensus-bad.json" <<'EOF'
+{
+  "accepted": [],
+  "deferred": [],
+  "rejected": [],
+  "approved": true,
+  "consensus": {"reviewerCount": 2, "verdict": "maybe"}
+}
+EOF
+check_via_validator "$TMP/consensus-bad.json" 1 "bad consensus verdict triggers exit 1"
+
+echo ""
+echo "→ fixture 10: consensus missing required reviewerCount (must reject  -  exit 1)"
+cat > "$TMP/consensus-noreqcount.json" <<'EOF'
+{
+  "accepted": [],
+  "deferred": [],
+  "rejected": [],
+  "approved": true,
+  "consensus": {"verdict": "unanimous-pass"}
+}
+EOF
+check_via_validator "$TMP/consensus-noreqcount.json" 1 "consensus without reviewerCount triggers exit 1"
+
+echo ""
+echo "→ fixture 11: consensus unanimous-block contradicts approved=true (exit 1)"
+cat > "$TMP/consensus-contra1.json" <<'EOF'
+{"accepted":[],"deferred":[],"rejected":[],"approved":true,"consensus":{"reviewerCount":2,"verdict":"unanimous-block"}}
+EOF
+check_via_validator "$TMP/consensus-contra1.json" 1 "unanimous-block + approved=true is rejected"
+
+echo ""
+echo "→ fixture 12: unanimous-pass contradicts an accepted blocker (exit 1)"
+cat > "$TMP/consensus-contra2.json" <<'EOF'
+{"accepted":[{"severity":"blocking","file":"a.ts","line":1,"issue":"sql injection here","fix":"parameterize the query","reviewer":"opus"}],"deferred":[],"rejected":[],"approved":false,"consensus":{"reviewerCount":2,"verdict":"unanimous-pass"}}
+EOF
+check_via_validator "$TMP/consensus-contra2.json" 1 "unanimous-pass + accepted blocker is rejected"
+
+echo ""
+echo "→ fixture 13: single reviewer cannot be unanimous (exit 1)"
+cat > "$TMP/consensus-solo.json" <<'EOF'
+{"accepted":[],"deferred":[],"rejected":[],"approved":true,"consensus":{"reviewerCount":1,"verdict":"unanimous-pass"}}
+EOF
+check_via_validator "$TMP/consensus-solo.json" 1 "reviewerCount 1 + unanimous verdict is rejected"
+
 echo ""
 echo "══ phase4-triage smoke: $PASS passed, $FAIL failed ══"
 [ "$FAIL" -eq 0 ]

package/pipeline/scripts/smoke-plan-approval-gate.sh

@@ -1,5 +1,5 @@
 #!/usr/bin/env bash
-# smoke-plan-approval-gate.sh — contract check for Phase 2 Plan Approval Gate (v5.3.0).
+# smoke-plan-approval-gate.sh  -  contract check for Phase 2 Plan Approval Gate (v5.3.0).
 #
 # Verifies:
 #   1. phase-2-planning.md declares Step 5 as the gate (clarification + approval)
@@ -33,7 +33,7 @@ grep -q "Clarification Mode" "$PHASE2" && pass "phase-2 has Clarification Mode (
 grep -q "Approval Loop" "$PHASE2" && pass "phase-2 has Approval Loop (5b)" || fail "phase-2 missing 5b"
 
 echo ""
-echo "→ 2. Gate is scoped — autopilot + --dev skip"
+echo "→ 2. Gate is scoped  -  autopilot + --dev skip"
 grep -q "state.autopilot === true" "$PHASE2" && pass "phase-2 guards against autopilot" || fail "phase-2 missing autopilot guard"
 grep -q "state.onlyDevelop === true" "$PHASE2" && pass "phase-2 guards against --dev" || fail "phase-2 missing --dev guard"
 
@@ -63,7 +63,7 @@ grep -qi "Gate skip" "$SKILL" && pass "SKILL.md autopilot matrix shows 'Gate ski
 
 echo ""
 echo "→ 7. Free-text edit contract (no sentinel required)"
-grep -qi "serbest metin" "$PHASE2" && pass "phase-2 documents free-text edit in TR" || fail "phase-2 missing 'serbest metin'"
+grep -qi "free-text" "$PHASE2" && pass "phase-2 documents the free-text edit path" || fail "phase-2 missing free-text edit path"
 grep -q 'planEditRequests' "$PHASE2" && pass "phase-2 references planEditRequests" || fail "phase-2 missing planEditRequests"
 
 echo ""

package/pipeline/scripts/smoke-plan-safety.sh

@@ -1,5 +1,5 @@
 #!/usr/bin/env bash
-# smoke-plan-safety.sh — v7.0.G
+# smoke-plan-safety.sh  -  v7.0.G
 #
 # Verifies the Phase 2 autopilot safety classifier:
 #   1. Low-risk plan: score low, recommendPause=false

package/pipeline/scripts/smoke-plan-todos.sh

@@ -52,7 +52,7 @@ for s in "${SUBS[@]}"; do
   elif bash "$HELPER" "$s" 2>&1 | grep -qE "usage: $s"; then
     record_pass "subcommand '$s' prints usage"
   else
-    # Some subcommands fail without args via exit 64 — that also confirms they're known
+    # Some subcommands fail without args via exit 64  -  that also confirms they're known
     bash "$HELPER" "$s" 2>/dev/null
     rc=$?
     if [ "$rc" = "64" ] || [ "$rc" = "1" ]; then
@@ -171,7 +171,7 @@ else
 fi
 
 # 10-12. Phase docs wiring
-if grep -qE 'Step 4\.5 — Emit Plan Todo List' "$P2"; then
+if grep -qE 'Step 4\.5  -  Emit Plan Todo List' "$P2"; then
   record_pass "phase-2-planning.md documents Step 4.5"
 else
   record_fail "phase-2-planning.md missing Step 4.5"

package/pipeline/scripts/smoke-pr-review-actions.sh

@@ -44,7 +44,7 @@ if [ -e "$LEGACY_RENDERER" ]; then
   echo "FAIL [smoke-pr-review-actions]: $LEGACY_RENDERER must not exist (monolith comment renderer forbidden)"
   FAIL=$((FAIL + 1))
 fi
-# Reference scan — no spec files should still WIRE the legacy artifacts.
+# Reference scan  -  no spec files should still WIRE the legacy artifacts.
 # Match only call-site patterns (bash <path>, --body-file <path>), NOT prose
 # references. Migration notes like "render-pr-review-body.sh is GONE" are
 # allowed; an actual `bash render-pr-review-body.sh` call is not.
@@ -73,7 +73,7 @@ for a in "${DECISION_ANCHORS[@]}"; do
   fi
 done
 
-# 4. Both languages render — the template documents tr + en shapes.
+# 4. Both languages render  -  the template documents tr + en shapes.
 if ! grep -qE 'outputLanguage *= *"tr"|Turkish \(outputLanguage' "$CHANNEL"; then
   echo "FAIL [smoke-pr-review-actions]: $CHANNEL must document the tr inline-comment shape"
   FAIL=$((FAIL + 1))

package/pipeline/scripts/smoke-pre-commit.sh

@@ -1,7 +1,7 @@
 #!/usr/bin/env bash
-# smoke-pre-commit.sh — verify pre-commit-check.sh catches all secret patterns.
+# smoke-pre-commit.sh  -  verify pre-commit-check.sh catches all secret patterns.
 # Uses a real temp git repo to test the actual hook behavior end-to-end.
-# No mocks, no test-only flags — tests exactly what runs in production.
+# No mocks, no test-only flags  -  tests exactly what runs in production.
 
 set -uo pipefail
 
@@ -96,6 +96,30 @@ check_detects "GCP service account JSON" \
   "test-file.json" \
   '{"type": "service_account", "project_id": "my-project"}'
 
+check_detects "GitHub PAT (ghp_ 36 chars)" \
+  "test-file.txt" \
+  'token: ghp_aBcDeFgHiJkLmNoPqRsTuVwXyZ0123456789'
+
+check_detects "Slack token (xoxb-)" \
+  "test-file.txt" \
+  'slack = "xoxb-123456789012-abcdefghijklmnop"'
+
+check_detects "Google API key (AIza)" \
+  "test-file.txt" \
+  'gmaps = "AIzaSyA1234567890abcdefghijklmnopqrstuv"'
+
+check_detects "JWT three-segment token" \
+  "test-file.txt" \
+  'auth = "eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.eyJzdWIiOiIxMjM0NTY3ODkwIn0.dozjgNryP4J3jVmNHl0w5N"'
+
+check_detects "High-entropy custom secret (no known prefix)" \
+  "test-file.txt" \
+  'session = "Zx9Kq2Lm7Pw4Nv8Rt3Yb6Hc1Df5Gj0Sa2Wd7Qe4Uo9Ik"'
+
+check_detects "Secret in a filename WITH SPACES (no false-negative)" \
+  "test secret file.txt" \
+  'token: ghp_aBcDeFgHiJkLmNoPqRsTuVwXyZ0123456789'
+
 echo ""
 echo "→ 2. File-name based detection"
 
@@ -124,6 +148,14 @@ check_passes "Empty file" \
   "test-file.txt" \
   ""
 
+check_passes "Git SHA / hex hash is not flagged as high-entropy" \
+  "test-file.txt" \
+  'commit a1b2c3d4e5f6789012345678901234567890abcd resolved'
+
+check_passes "Lockfile integrity hash is exempt" \
+  "test-yarn.lock" \
+  'resolved "https://r.npm/x" integrity sha512-Zx9Kq2Lm7Pw4Nv8Rt3Yb6Hc1Df5Gj0Sa2Wd7Qe4Uo9IkLp=='
+
 echo ""
 echo "→ 4. No staged files = clean exit"
 reset_stage

package/pipeline/scripts/smoke-pref-migration.sh

@@ -1,5 +1,5 @@
 #!/usr/bin/env bash
-# smoke-pref-migration.sh — exercises migrate-prefs.mjs against representative
+# smoke-pref-migration.sh  -  exercises migrate-prefs.mjs against representative
 # v2.0.0 and pre-2.0.0 fixtures, asserts the output is v2.1.0-shaped.
 #
 # Fixtures are synthesized inline (no external files) to keep the test self-contained.

package/pipeline/scripts/smoke-prefs-language.sh

@@ -1,5 +1,5 @@
 #!/usr/bin/env bash
-# smoke-prefs-language.sh — contract check for `/multi-agent language`.
+# smoke-prefs-language.sh  -  contract check for `/multi-agent language`.
 #
 # Does NOT touch the user's real $HOME/.claude/multi-agent-preferences.json.
 # Uses a throwaway tempfile + node shim that applies the documented write
@@ -116,12 +116,12 @@ TCC_PCT_MAX=$(jq -r '.properties.global.properties.triageCrossCheck.properties.s
 [ "$TCC_PCT_MIN" = "1" ] && [ "$TCC_PCT_MAX" = "100" ] && pass "samplePct bounded [1,100]" || fail "samplePct bounds wrong"
 
 echo ""
-echo "→ setup.md documents Step 0 — Language Selection"
+echo "→ setup.md documents Step 0  -  Language Selection"
 SETUP_DOC="$ROOT/pipeline/commands/multi-agent/setup.md"
-if [ -f "$SETUP_DOC" ] && grep -q "^### Step 0 — Language Selection" "$SETUP_DOC"; then
-  pass "setup.md has Step 0 — Language Selection"
+if [ -f "$SETUP_DOC" ] && grep -q "^### Step 0  -  Language Selection" "$SETUP_DOC"; then
+  pass "setup.md has Step 0  -  Language Selection"
 else
-  fail "setup.md missing Step 0 — Language Selection heading"
+  fail "setup.md missing Step 0  -  Language Selection heading"
 fi
 if [ -f "$SETUP_DOC" ] && grep -q "promptLanguage" "$SETUP_DOC"; then
   pass "setup.md references promptLanguage"

package/pipeline/scripts/smoke-progress-contract.sh

@@ -1,5 +1,5 @@
 #!/usr/bin/env bash
-# smoke-progress-contract.sh — enforce the progress-line contract across phase docs.
+# smoke-progress-contract.sh  -  enforce the progress-line contract across phase docs.
 #
 # Verifies:
 #   1. Contract doc exists at refs/progress-contract.md with the required sections.
@@ -89,7 +89,7 @@ else
   if [ "$BLOCKING_ADOPTION" = "true" ]; then
     fail "missing adoption in ${#MISSING[@]} phases: ${MISSING[*]}"
   else
-    warn "advisory — missing adoption in ${#MISSING[@]} phases: ${MISSING[*]}"
+    warn "advisory  -  missing adoption in ${#MISSING[@]} phases: ${MISSING[*]}"
   fi
 fi
 
@@ -110,7 +110,7 @@ for doc in "$PHASE_DIR"/phase-*.md; do
   fi
 done
 if [ "$ADOPTED" -eq 0 ]; then
-  warn "no phases have adopted yet — this is expected until WS-4/5/6 land"
+  warn "no phases have adopted yet  -  this is expected until WS-4/5/6 land"
 fi
 
 echo ""

package/pipeline/scripts/smoke-prune-logs.sh

@@ -0,0 +1,87 @@
+#!/usr/bin/env bash
+# smoke-prune-logs.sh  -  prune-logs.sh removes per-task log dirs, honors the
+# project/task/age filters, is dry-run by default, and NEVER deletes the audit
+# trail or metrics corpus.
+#
+# Exit 0 = all pass, 1 = any failure.
+
+set -uo pipefail
+
+ROOT="$(cd "$(dirname "$0")/../.." && pwd)"
+PL="$ROOT/pipeline/scripts/prune-logs.sh"
+PASS=0
+FAIL=0
+pass() {
+  PASS=$((PASS + 1))
+  echo "  ✓ $1"
+}
+fail() {
+  FAIL=$((FAIL + 1))
+  echo "  ✗ $1"
+}
+[ -f "$PL" ] || {
+  echo "FAIL: prune-logs.sh missing" >&2
+  exit 1
+}
+
+LR="$(mktemp -d)"
+trap 'rm -rf "$LR"' EXIT
+
+seed() {
+  rm -rf "${LR:?}"/*
+  mkdir -p "$LR/my-ios-app/PROJ-1" "$LR/my-ios-app/PROJ-2" "$LR/other/PROJ-9"
+  echo '{}' >"$LR/my-ios-app/PROJ-1/agent-state.json"
+  echo '{}' >"$LR/my-ios-app/PROJ-2/tracker-state.json"
+  echo '{}' >"$LR/other/PROJ-9/agent-log.md"
+  echo 'audit' >"$LR/audit.jsonl"
+  echo 'metric' >"$LR/metrics.jsonl"
+  echo '5' >"$LR/my-ios-app/.counter"
+}
+
+echo "→ 1. dry-run lists task dirs and deletes nothing"
+seed
+out=$(PRUNE_LOG_ROOT="$LR" bash "$PL" 2>&1)
+echo "$out" | grep -q "DRY-RUN" && pass "dry-run banner shown" || fail "dry-run banner missing"
+[ -f "$LR/my-ios-app/PROJ-1/agent-state.json" ] && pass "dry-run kept task dirs" || fail "dry-run deleted a task dir"
+echo "$out" | grep -qi "preserved" && pass "dry-run states audit/metrics are preserved" || fail "no preservation note"
+
+echo ""
+echo "→ 2. --yes deletes task dirs but preserves audit + metrics + .counter"
+seed
+PRUNE_LOG_ROOT="$LR" bash "$PL" --yes >/dev/null 2>&1
+[ -f "$LR/audit.jsonl" ] && pass "audit.jsonl preserved" || fail "audit.jsonl was deleted"
+[ -f "$LR/metrics.jsonl" ] && pass "metrics.jsonl preserved" || fail "metrics.jsonl was deleted"
+[ -f "$LR/my-ios-app/.counter" ] && pass ".counter preserved" || fail ".counter was deleted"
+[ ! -e "$LR/my-ios-app/PROJ-1" ] && pass "task dir PROJ-1 removed" || fail "task dir PROJ-1 survived"
+[ ! -e "$LR/other/PROJ-9" ] && pass "task dir PROJ-9 removed" || fail "task dir PROJ-9 survived"
+
+echo ""
+echo "→ 3. --project scopes deletion to one project"
+seed
+PRUNE_LOG_ROOT="$LR" bash "$PL" --project=my-ios-app --yes >/dev/null 2>&1
+[ ! -e "$LR/my-ios-app/PROJ-1" ] && pass "my-ios-app task removed" || fail "my-ios-app task survived"
+[ -f "$LR/other/PROJ-9/agent-log.md" ] && pass "other project untouched" || fail "other project was deleted"
+
+echo ""
+echo "→ 4. --task scopes deletion to one task dir"
+seed
+PRUNE_LOG_ROOT="$LR" bash "$PL" --task=PROJ-1 --yes >/dev/null 2>&1
+[ ! -e "$LR/my-ios-app/PROJ-1" ] && pass "PROJ-1 removed" || fail "PROJ-1 survived"
+[ -f "$LR/my-ios-app/PROJ-2/tracker-state.json" ] && pass "PROJ-2 untouched" || fail "PROJ-2 was deleted"
+
+echo ""
+echo "→ 5. --older-than excludes fresh task dirs"
+seed
+out=$(PRUNE_LOG_ROOT="$LR" bash "$PL" --older-than=1 2>&1)
+echo "$out" | grep -q "nothing" && pass "freshly-created dirs excluded by --older-than=1" || fail "--older-than did not exclude fresh dirs"
+
+echo ""
+echo "→ 6. usage errors + empty root"
+PRUNE_LOG_ROOT="$LR" bash "$PL" --bogus >/dev/null 2>&1
+[ "$?" = "2" ] && pass "unknown argument exits 2" || fail "unknown argument did not exit 2"
+PRUNE_LOG_ROOT="/nonexistent-$$-dir" bash "$PL" >/dev/null 2>&1
+[ "$?" = "0" ] && pass "missing log root exits 0 (no-op)" || fail "missing log root did not exit 0"
+
+echo ""
+echo "══ prune-logs smoke: $PASS passed, $FAIL failed ══"
+[ "$FAIL" -eq 0 ]

package/pipeline/scripts/smoke-push-retry.sh

@@ -1,5 +1,5 @@
 #!/usr/bin/env bash
-# smoke-push-retry.sh — asserts the push-must-succeed policy is documented
+# smoke-push-retry.sh  -  asserts the push-must-succeed policy is documented
 # correctly in Phase 6 and gated by settings.pushMustSucceed.
 
 set -uo pipefail

package/pipeline/scripts/smoke-readme-counts.sh

@@ -1,5 +1,5 @@
 #!/usr/bin/env bash
-# smoke-readme-counts.sh — fail when README "Current at-a-glance" counts drift
+# smoke-readme-counts.sh  -  fail when README "Current at-a-glance" counts drift
 # from the actual filesystem. The table is hand-edited; without this gate it
 # silently goes stale every time a smoke / skill / schema is added or removed.
 #

package/pipeline/scripts/smoke-repo-map.sh

@@ -10,8 +10,8 @@
 #   6. Python fixture extracts class/def
 #   7. Bash fixture extracts top-level functions
 #   8. TF-IDF rule: an identifier defined in many files does not dominate rank
-#   9. Token budget honored — output stops at budget
-#  10. Determinism — same input twice = byte-identical output
+#   9. Token budget honored  -  output stops at budget
+#  10. Determinism  -  same input twice = byte-identical output
 #  11. JSON format emits valid JSON with score / declarations
 #  12. Word-boundary match: substring of an identifier is NOT counted
 #
@@ -184,21 +184,21 @@ else
   record_fail "Bash extractor missed functions"
 fi
 
-# --- 8: TF-IDF rule — rare identifier ranks higher than shared identifier ---
+# --- 8: TF-IDF rule  -  rare identifier ranks higher than shared identifier ---
 # OrderViewModel defined once, referenced from refs.swift.
 # CommonName defined twice (A + B), referenced from refs.swift.
 # refs.swift's outgoing edge to Order.swift should weigh more than to A/B
-# combined — so Order.swift should rank above A.swift in this fixture.
+# combined  -  so Order.swift should rank above A.swift in this fixture.
 order_line=$(echo "$MD" | grep -nF "Order.swift" | head -1 | cut -d: -f1)
 a_line=$(echo "$MD" | grep -nF "shared/A.swift" | head -1 | cut -d: -f1)
 if [ -n "$order_line" ] && [ -n "$a_line" ]; then
   if [ "$order_line" -lt "$a_line" ]; then
-    record_pass "TF-IDF — rarely-defined identifier outranks shared identifier"
+    record_pass "TF-IDF  -  rarely-defined identifier outranks shared identifier"
   else
     record_fail "TF-IDF rule failed: Order.swift should rank above shared/A.swift"
   fi
 else
-  record_fail "TF-IDF — could not locate Order.swift or shared/A.swift in output"
+  record_fail "TF-IDF  -  could not locate Order.swift or shared/A.swift in output"
 fi
 
 # --- 9: token budget honored -----------------------------------------------
@@ -215,7 +215,7 @@ fi
 RUN1=$(node "$SCRIPT" --root "$FIX" --budget 800 --top 10 --format md 2>&1)
 RUN2=$(node "$SCRIPT" --root "$FIX" --budget 800 --top 10 --format md 2>&1)
 if [ "$RUN1" = "$RUN2" ]; then
-  record_pass "determinism — same input twice produces byte-identical output"
+  record_pass "determinism  -  same input twice produces byte-identical output"
 else
   record_fail "non-deterministic output across runs"
 fi
@@ -252,7 +252,7 @@ for d in data:
         break
 ')
 if [ -n "$def_refs" ] && python3 -c "import sys; sys.exit(0 if abs(${def_refs}) < 0.01 else 1)" 2>/dev/null; then
-  record_pass "word-boundary — super-string match (WidgetXYZ123Helper) does not count as WidgetXYZ123 ref"
+  record_pass "word-boundary  -  super-string match (WidgetXYZ123Helper) does not count as WidgetXYZ123 ref"
 else
   record_fail "word-boundary regression: def.swift in_refs=$def_refs (expected ~0)"
 fi
@@ -271,7 +271,7 @@ for prop in tokenBudget topFiles include exclude; do
     record_fail "prefs schema missing repoMap.${prop}"
   fi
 done
-# Off by default — preserve baseline for existing users.
+# Off by default  -  preserve baseline for existing users.
 if jq -e '.properties.global.properties.repoMap.properties.enabled
           | has("default") and .default == false' "$SCHEMA_PATH" >/dev/null 2>&1; then
   record_pass "repoMap.enabled defaults to false (opt-in)"

package/pipeline/scripts/smoke-review-watch.sh

@@ -125,6 +125,18 @@ else
   record_pass "unknown command exits non-zero"
 fi
 
+# 11. PR loop must NOT pipe into `while` (a pipe subshell discards every
+# state= / cursor mutation, so save_state would persist the stale cursor and
+# the watcher would re-dispatch the same PRs forever). It must use process
+# substitution so the loop runs in the function's own shell.
+if grep -qE "jq -c '\.\[\]' <<<\"\\\$prs_json\" \| while" "$WATCH"; then
+  record_fail "review-watch PR loop pipes into while (subshell loses cursor state)"
+elif awk '/while read -r pr;/,/done < <\(jq/' "$WATCH" | grep -q 'done < <(jq'; then
+  record_pass "PR loop uses process substitution (cursor mutations persist)"
+else
+  record_fail "PR loop structure unrecognized  -  verify it does not pipe into while"
+fi
+
 printf '\n══ review-watch smoke: %d passed, %d failed ══\n' "$pass" "$fail"
 if [ "$fail" -gt 0 ]; then
   printf '\nFailures:\n'

package/pipeline/scripts/smoke-run-aggregator.sh

@@ -1,5 +1,5 @@
 #!/usr/bin/env bash
-# smoke-run-aggregator.sh — v7.8.0 Paket C: unified telemetry aggregator.
+# smoke-run-aggregator.sh  -  v7.8.0 Paket C: unified telemetry aggregator.
 #
 # Validates run-aggregator.mjs reads phase-tracker + otel-spans + cost table
 # and emits a normalized pipeline-run.jsonl with:
@@ -68,7 +68,7 @@ else
 fi
 
 # ──────────────────────────────────────────────────────────────────────────
-echo "→ 3. Read-only contract — source files untouched"
+echo "→ 3. Read-only contract  -  source files untouched"
 HASH_AFTER=$(md5 -q "$TMP/phase-tracker.json" "$TMP/otel-spans.jsonl" 2>/dev/null || \
              md5sum "$TMP/phase-tracker.json" "$TMP/otel-spans.jsonl" 2>/dev/null | awk '{print $1}' | tr -d '\n')
 if [ "$HASH_BEFORE" = "$HASH_AFTER" ]; then
@@ -97,15 +97,15 @@ fi
 # ──────────────────────────────────────────────────────────────────────────
 echo "→ 5. Cost computation correct"
 # sonnet 1000+200 = 0.001*3 + 0.0002*15 = 0.003 + 0.003 = 0.006
-# opus 50000+3000 = 0.05*15 + 0.003*75 = 0.75 + 0.225 = 0.975
+# opus 50000+3000 = 0.05*5 + 0.003*25 = 0.25 + 0.075 = 0.325
 # sonnet 10000+2000 = 0.01*3 + 0.002*15 = 0.03 + 0.03 = 0.060
-# total = 0.006 + 0.975 + 0.060 = 1.041
+# total = 0.006 + 0.325 + 0.060 = 0.391
 SUMMARY=$(tail -1 "$TMP/pipeline-run.jsonl")
 TOTAL=$(echo "$SUMMARY" | grep -oE '"total_cost_usd":[^,}]*' | sed 's/"total_cost_usd"://')
-if [ "$TOTAL" = "1.041" ]; then
+if [ "$TOTAL" = "0.391" ]; then
   pass "total_cost_usd = $TOTAL (matches expected)"
 else
-  fail "total_cost_usd = $TOTAL (expected 1.041)"
+  fail "total_cost_usd = $TOTAL (expected 0.391)"
 fi
 
 # ──────────────────────────────────────────────────────────────────────────
@@ -189,7 +189,7 @@ else
 fi
 
 # ──────────────────────────────────────────────────────────────────────────
-echo "→ 12. Idempotent — second run produces identical content"
+echo "→ 12. Idempotent  -  second run produces identical content"
 # Restore the first fixture
 cat > "$TMP/phase-tracker.json" <<'EOF'
 {

package/pipeline/scripts/smoke-run-metrics.sh

@@ -0,0 +1,50 @@
+#!/usr/bin/env bash
+# smoke-run-metrics.sh  -  run-metrics.mjs extracts the right outcome numbers
+# from an agent-state.json. This is the evidence-collection harness: run it
+# after each real task and the corpus becomes the real-world validation.
+#
+# Exit 0 = all pass, 1 = any failure.
+
+set -uo pipefail
+
+ROOT="$(cd "$(dirname "$0")/../.." && pwd)"
+M="$ROOT/pipeline/scripts/run-metrics.mjs"
+FIX="$ROOT/pipeline/eval/run-metrics-fixture.json"
+PASS=0; FAIL=0
+pass() { PASS=$((PASS+1)); echo "  ✓ $1"; }
+fail() { FAIL=$((FAIL+1)); echo "  ✗ $1"; }
+command -v node >/dev/null 2>&1 || { echo "error: node required" >&2; exit 127; }
+command -v jq >/dev/null 2>&1 || { echo "error: jq required" >&2; exit 127; }
+[ -f "$M" ] || { echo "FAIL: run-metrics.mjs missing" >&2; exit 1; }
+[ -f "$FIX" ] || { echo "FAIL: fixture missing" >&2; exit 1; }
+
+out=$(node "$M" --state "$FIX" 2>/dev/null)
+
+check() { # jq-expr desc
+  if jq -e "$1" >/dev/null 2>&1 <<< "$out"; then pass "$2"; else fail "$2 (got: $(jq -c "$3" <<< "$out" 2>/dev/null))"; fi
+}
+
+echo "→ 1. core outcome fields"
+check '.status == "complete"' "status parsed" '.status'
+check '.reviewIterations == 2' "review iterations counted (rework loop)" '.reviewIterations'
+check '.firstPassClean == false' "first-pass-clean detects iteration-1 blocker" '.firstPassClean'
+check '.buildPassed == true' "build outcome aggregated across repos" '.buildPassed'
+
+echo ""
+echo "→ 2. review signal-to-noise"
+check '.reviewSignal.rawFindings == 3' "raw reviewer findings summed" '.reviewSignal.rawFindings'
+check '.reviewSignal.acceptedAll == 1' "accepted summed across iterations" '.reviewSignal.acceptedAll'
+check '.reviewSignal.acceptedRatio == 0.33' "accepted ratio (signal-to-noise)" '.reviewSignal.acceptedRatio'
+
+echo ""
+echo "→ 3. consensus surfaced"
+check '.consensusVerdict == "unanimous-pass"' "consensus verdict extracted" '.consensusVerdict'
+
+echo ""
+echo "→ 4. stdin + missing-field resilience"
+echo '{"taskId":"X","status":"failed"}' | node "$M" --state - >/dev/null 2>&1 && pass "minimal state via stdin does not crash" || fail "stdin/minimal state crashed"
+node "$M" 2>/dev/null; [ "$?" = "2" ] && pass "no --state -> usage exit 2" || fail "missing --state wrong exit"
+
+echo ""
+echo "══ run-metrics smoke: $PASS passed, $FAIL failed ══"
+[ "$FAIL" -eq 0 ]