@mmerterden/multi-agent-pipeline 8.6.0

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

@@ -0,0 +1,223 @@
+#!/usr/bin/env bash
+# smoke-bitbucket-contract.sh — offline contract test for repo-cache.sh
+# Bitbucket Server backend.
+#
+# Replaces curl with a stub that serves a fixture JSON, then verifies the
+# python parser shape-converts the BB Server `/rest/api/1.0/projects/.../repos`
+# response into the canonical {fullName, name, owner, cloneUrl, provider}
+# schema that the picker consumes.
+#
+# Why offline: live Bitbucket Server REST requires a corporate host + project
+# key + valid token. Without those, contract drift would only surface in
+# production. This smoke pins the request URL, the response parsing, and the
+# auth header construction so changes that break either fail in CI.
+
+set -uo pipefail
+
+SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd)"
+REPO_ROOT="$(cd "$SCRIPT_DIR/../.." && pwd)"
+LIB_DIR="$REPO_ROOT/pipeline/lib"
+
+PASS=0; FAIL=0
+ok()   { echo "  ✓ $1"; PASS=$((PASS+1)); }
+fail() { echo "  ✗ $1" >&2; FAIL=$((FAIL+1)); }
+
+REPO_CACHE="$LIB_DIR/repo-cache.sh"
+[ -x "$REPO_CACHE" ] || { fail "repo-cache.sh missing/not executable"; exit 1; }
+
+# --- Stub setup -------------------------------------------------------------
+TMP=$(mktemp -d)
+trap 'rm -rf "$TMP"' EXIT
+
+# Fixture: minimal but realistic BB Server response shape (as documented at
+# https://docs.atlassian.com/bitbucket-server/rest/7.21.0/bitbucket-rest.html)
+cat > "$TMP/fixture.json" <<'EOF'
+{
+  "size": 2,
+  "limit": 200,
+  "isLastPage": true,
+  "values": [
+    {
+      "slug": "my-ios-app",
+      "name": "my-ios-app",
+      "project": {"key": "MOBILE", "name": "Mobile Team"},
+      "description": "iOS application",
+      "links": {
+        "clone": [
+          {"href": "https://bitbucket.example.com/scm/mobile/my-ios-app.git", "name": "http"},
+          {"href": "ssh://git@bitbucket.example.com/mobile/my-ios-app.git", "name": "ssh"}
+        ]
+      }
+    },
+    {
+      "slug": "common-ui-kit",
+      "name": "common-ui-kit",
+      "project": {"key": "MOBILE", "name": "Mobile Team"},
+      "description": "Shared UI components",
+      "links": {
+        "clone": [
+          {"href": "https://bitbucket.example.com/scm/mobile/common-ui-kit.git", "name": "http"},
+          {"href": "ssh://git@bitbucket.example.com/mobile/common-ui-kit.git", "name": "ssh"}
+        ]
+      }
+    }
+  ]
+}
+EOF
+
+# Stub curl that captures invocation and returns the fixture. PATH-shadows the
+# real curl for this smoke only.
+cat > "$TMP/curl" <<EOF
+#!/usr/bin/env bash
+# Capture every invocation so we can assert the URL + auth header shape.
+echo "INVOCATION:" "\$@" >> "$TMP/curl-calls.log"
+# Echo the fixture body (BB Server response) to stdout — same as real curl -sf.
+cat "$TMP/fixture.json"
+exit 0
+EOF
+chmod +x "$TMP/curl"
+
+# Stub `security` so credential-store.sh `get` returns deterministic test values.
+# We do this rather than touching the real Keychain.
+cat > "$TMP/security" <<EOF
+#!/usr/bin/env bash
+# Mimic the relevant security subcommands used by credential-store.sh.
+case "\$1" in
+  find-generic-password)
+    # macOS path. Args: -s NAME -w
+    while [ \$# -gt 0 ]; do
+      case "\$1" in
+        -s) shift; SVC="\$1" ;;
+      esac
+      shift || break
+    done
+    case "\$SVC" in
+      mock_BB_Token) echo "test-bb-token" ;;
+      mock_BB_User)  echo "test-bb-user" ;;
+      *) exit 44 ;;
+    esac
+    ;;
+  dump-keychain) ;; # noop for list
+esac
+EOF
+chmod +x "$TMP/security"
+
+# Override repo-cache TTL and stub PATH so curl + security come from $TMP.
+export PATH="$TMP:$PATH"
+export REPO_CACHE_TTL=0   # skip cache, force fetch
+export BB_TOKEN_KEY="mock_BB_Token"
+export BB_USER_KEY="mock_BB_User"
+export BB_HOST="bitbucket.example.com"
+
+# Run with a sandboxed cache dir so we don't pollute the real one
+export HOME="$TMP/fakehome"
+mkdir -p "$HOME/.claude/cache/repos" "$HOME/.claude/lib"
+ln -sf "$LIB_DIR/credential-store.sh" "$HOME/.claude/lib/credential-store.sh"
+
+# --- Run the call -----------------------------------------------------------
+echo "→ 1. repo-cache.sh bitbucket MOBILE — invoke + parse"
+out=$("$REPO_CACHE" bitbucket MOBILE --refresh 2>&1)
+if [ $? -eq 0 ]; then
+  ok "exit 0 with valid fixture"
+else
+  fail "exit non-zero on valid fixture: $out"
+fi
+
+# --- Assert: curl called with correct URL + auth ----------------------------
+echo "→ 2. curl invocation captured the right URL"
+if grep -q "https://bitbucket.example.com/rest/api/1.0/projects/MOBILE/repos" "$TMP/curl-calls.log"; then
+  ok "URL matches BB Server REST contract (/rest/api/1.0/projects/{key}/repos)"
+else
+  fail "URL wrong; captured: $(cat "$TMP/curl-calls.log")"
+fi
+
+if grep -q "limit=200" "$TMP/curl-calls.log"; then
+  ok "page limit=200 (the BB Server max)"
+else
+  fail "limit param missing"
+fi
+
+if grep -q "test-bb-user:test-bb-token" "$TMP/curl-calls.log"; then
+  ok "Basic auth uses {user}:{token} from credential store"
+else
+  fail "auth header wrong; expected user:token, got: $(grep -o '\-u [^ ]*' "$TMP/curl-calls.log" || echo none)"
+fi
+
+# --- Assert: shape conversion produced canonical schema ---------------------
+echo "→ 3. parser converts BB shape → canonical schema"
+canonical=$(printf '%s' "$out" | python3 -c '
+import json, sys
+data = json.load(sys.stdin)
+required = {"fullName","name","owner","description","isFork","updatedAt","cloneUrl","provider"}
+for r in data:
+    miss = required - set(r.keys())
+    if miss:
+        sys.stderr.write("MISS:" + str(sorted(miss)) + "\n")
+        sys.exit(1)
+    if r["provider"] != "bitbucket":
+        sys.stderr.write("BAD_PROVIDER:" + r["provider"] + "\n")
+        sys.exit(1)
+print(json.dumps(data))
+')
+if [ -n "$canonical" ]; then
+  ok "every output entry has the canonical fields + provider=bitbucket"
+else
+  fail "shape conversion failed"
+fi
+
+count=$(printf '%s' "$canonical" | python3 -c 'import json,sys; print(len(json.load(sys.stdin)))')
+if [ "$count" = "2" ]; then
+  ok "all values[] entries surfaced (count matches fixture)"
+else
+  fail "count $count != 2"
+fi
+
+# Specific field plumbing
+first_full=$(printf '%s' "$canonical" | python3 -c 'import json,sys; print(json.load(sys.stdin)[0]["fullName"])')
+if [ "$first_full" = "MOBILE/my-ios-app" ]; then
+  ok "fullName format: <projectKey>/<slug>"
+else
+  fail "fullName wrong: $first_full"
+fi
+
+first_clone=$(printf '%s' "$canonical" | python3 -c 'import json,sys; print(json.load(sys.stdin)[0]["cloneUrl"])')
+if printf '%s' "$first_clone" | grep -q "^ssh://"; then
+  ok "cloneUrl prefers ssh link when both available"
+else
+  fail "cloneUrl should prefer ssh, got: $first_clone"
+fi
+
+# --- Assert: empty values[] handled cleanly ---------------------------------
+echo "→ 4. empty BB project (no repos) returns empty array"
+cat > "$TMP/fixture.json" <<'EOF'
+{"size": 0, "limit": 200, "isLastPage": true, "values": []}
+EOF
+out=$("$REPO_CACHE" bitbucket EMPTY --refresh 2>&1)
+parsed=$(printf '%s' "$out" | python3 -c 'import json,sys; print(len(json.load(sys.stdin)))' 2>/dev/null || echo "ERR")
+if [ "$parsed" = "0" ]; then
+  ok "empty project → []"
+else
+  fail "empty project parse failed (got: $parsed)"
+fi
+
+# --- Assert: missing token exits with diagnostic ----------------------------
+echo "→ 5. missing BB token → exit 3 with hint"
+unset BB_TOKEN_KEY
+err=$("$REPO_CACHE" bitbucket MOBILE --refresh 2>&1 >/dev/null) || true
+# repo-cache.sh fails with `BB_TOKEN_KEY required` from `: "${BB_TOKEN_KEY:?...}"`
+if printf '%s' "$err" | grep -q "BB_TOKEN_KEY"; then
+  ok "missing BB_TOKEN_KEY produces a diagnostic"
+else
+  fail "no diagnostic on missing token (got: $err)"
+fi
+
+# --- Summary ----------------------------------------------------------------
+echo ""
+TOTAL=$((PASS+FAIL))
+if [ $FAIL -eq 0 ]; then
+  echo "══ bitbucket-contract smoke: $PASS passed, 0 failed ══"
+  exit 0
+else
+  echo "══ bitbucket-contract smoke: $PASS passed, $FAIL failed (of $TOTAL) ══" >&2
+  exit 1
+fi

package/pipeline/scripts/smoke-channels-flow.sh

@@ -0,0 +1,130 @@
+#!/usr/bin/env bash
+# smoke-channels-flow.sh — asserts the v5.7+ /multi-agent:channels contract:
+#   1. channels.md exists (Claude) + multi-agent-channels SKILL.md exists (Copilot)
+#   2. enrich.md is GONE (Claude) + multi-agent-enrich skill is GONE (Copilot)
+#   3. channels.md declares the 4 channels (pr, jira, confluence, wiki)
+#   4. channels.md declares the 4 content sources (normalAnalysis, testScenarios, auto-diff, manual-note)
+#   5. Reviewer-preserving Bitbucket PUT payload is present (title+description+reviewers+fromRef+toRef+version)
+#   6. 30-min timeout contract referenced in channels.md + phase-7-report.md + modes.md
+#   7. Phase 7 autopilot always-pause rule documented in modes.md
+#   8. Phase 7 report delegates to channels (does NOT inline Jira/Confluence/Wiki logic)
+#   9. schema has reportChannels, reportContent, autopilotReportTimeoutSeconds, wikiScope
+#  10. legacy jiraCommentDefault/confluenceDefault/wikiDefault REMOVED from schema
+#  11. migrate-prefs.mjs has migrateReportChannels() sub-migration
+#  12. preferences-template.json uses new fields (not legacy)
+#
+# Fails loudly on any violation. Intended to run as part of the full smoke suite.
+
+set -uo pipefail
+
+ROOT="$(cd "$(dirname "$0")/../.." && pwd)"
+CLAUDE_CMD="$ROOT/pipeline/commands/multi-agent/channels.md"
+COPILOT_SKILL="$ROOT/pipeline/skills/shared/core/multi-agent-channels/SKILL.md"
+ENRICH_CMD="$ROOT/pipeline/commands/multi-agent/enrich.md"
+ENRICH_SKILL_DIR="$ROOT/pipeline/skills/shared/core/multi-agent-enrich"
+PHASE7="$ROOT/pipeline/commands/multi-agent/refs/phases/phase-7-report.md"
+MODES="$ROOT/pipeline/commands/multi-agent/refs/phases/modes.md"
+WIKI_CAPTURE="$ROOT/pipeline/commands/multi-agent/refs/wiki-capture.md"
+SCHEMA="$ROOT/pipeline/schemas/prefs.schema.json"
+MIGRATE="$ROOT/pipeline/scripts/migrate-prefs.mjs"
+TEMPLATE="$ROOT/pipeline/preferences-template.json"
+
+PASS=0
+FAIL=0
+pass() { PASS=$((PASS + 1)); echo "  ✓ $1"; }
+fail() { FAIL=$((FAIL + 1)); echo "  ✗ $1"; }
+
+check() {
+  local label="$1" cmd="$2"
+  if eval "$cmd" > /dev/null 2>&1; then pass "$label"; else fail "$label"; fi
+}
+
+echo "→ 1. channels command exists (both CLIs)"
+check "Claude channels.md present"              "[ -f '$CLAUDE_CMD' ]"
+check "Copilot multi-agent-channels SKILL.md"   "[ -f '$COPILOT_SKILL' ]"
+
+echo ""
+echo "→ 2. enrich command + skill REMOVED"
+check "Claude enrich.md absent"                 "[ ! -f '$ENRICH_CMD' ]"
+check "Copilot multi-agent-enrich dir absent"   "[ ! -d '$ENRICH_SKILL_DIR' ]"
+
+echo ""
+echo "→ 3. channels.md declares 4 kanals"
+check "mentions PR description"                 "grep -Fq 'PR description' '$CLAUDE_CMD'"
+check "mentions Jira comment"                   "grep -Fq 'Jira comment'   '$CLAUDE_CMD'"
+check "mentions Confluence page"                "grep -Fq 'Confluence'     '$CLAUDE_CMD'"
+check "mentions Wiki (component-only)"          "grep -Fq 'Wiki'           '$CLAUDE_CMD'"
+
+echo ""
+echo "→ 4. channels.md declares 5 content sources (v5.7.3+ adds teknik analiz)"
+check "Normal analiz content option"            "grep -Fq 'Normal analiz'    '$CLAUDE_CMD'"
+check "Teknik analiz content option"            "grep -Fq 'Teknik analiz'    '$CLAUDE_CMD'"
+check "Test senaryoları content option"         "grep -Fq 'Test senaryoları' '$CLAUDE_CMD'"
+check "Auto-diff content option"                "grep -Fq 'Auto-diff'        '$CLAUDE_CMD'"
+check "Manuel not content option"               "grep -Fq 'Manuel not'       '$CLAUDE_CMD'"
+
+echo ""
+echo "→ 5. Reviewer-preserving Bitbucket PUT payload"
+check "mentions reviewers field mandatory"      "grep -Fq 'reviewers' '$CLAUDE_CMD'"
+check "mentions fromRef field mandatory"        "grep -Fq 'fromRef'   '$CLAUDE_CMD'"
+check "mentions toRef field mandatory"          "grep -Fq 'toRef'     '$CLAUDE_CMD'"
+check "mentions version field mandatory"        "grep -Fq 'version'   '$CLAUDE_CMD'"
+check "jq --rawfile pattern (real newlines)"    "grep -Eq 'jq --rawfile|jq -n' '$CLAUDE_CMD'"
+
+echo ""
+echo "→ 6. 30-min timeout contract"
+check "channels.md mentions 30-min timeout"     "grep -Eq '30.min|30-min|30 min' '$CLAUDE_CMD'"
+check "phase-7-report.md mentions timeout"      "grep -Eq '30.min|30-min|30 min|channelsTimeout' '$PHASE7'"
+check "modes.md mentions timeout"               "grep -Eq '30.min|30-min|30 min|timeout' '$MODES'"
+
+echo ""
+echo "→ 7. Phase 7 autopilot always-pause rule"
+check "modes.md: Phase 7 autopilot exception"   "grep -Fq 'Phase 7 autopilot exception' '$MODES'"
+check "modes.md: STILL PAUSES"                  "grep -Fq 'STILL PAUSES' '$MODES'"
+check "phase-7-report.md: always pauses"        "grep -Eq 'ALWAYS pauses|always pauses' '$PHASE7'"
+
+echo ""
+echo "→ 8. Phase 7 delegates to channels (no inline adapters)"
+check "phase-7-report.md invokes channels"      "grep -Eq 'multi-agent:channels|multi-agent-channels' '$PHASE7'"
+check "phase-7-report.md has state bundle"      "grep -Fq 'state bundle' '$PHASE7'"
+# Old inline per-channel sub-step list (Jira-Comment/Wiki+Figma/Confluence) should be gone
+if grep -Fq 'Jira-Comment' "$PHASE7" && grep -Fq 'Wiki+Figma' "$PHASE7" && grep -Fq 'Confluence' "$PHASE7"; then
+  # All three together = old 5-sub-step inline layout
+  fail "phase-7-report.md still has old inline 5-sub-step layout (Jira-Comment/Wiki+Figma/Confluence all three)"
+else
+  pass "phase-7-report.md no longer inlines old 5-sub-step layout"
+fi
+
+echo ""
+echo "→ 9. schema has new v5.7+ fields"
+check "schema has reportChannels"               "grep -Fq 'reportChannels'              '$SCHEMA'"
+check "schema has reportContent"                "grep -Fq 'reportContent'               '$SCHEMA'"
+check "schema has autopilotReportTimeoutSeconds" "grep -Fq 'autopilotReportTimeoutSeconds' '$SCHEMA'"
+check "schema has wikiScope"                    "grep -Fq 'wikiScope'                    '$SCHEMA'"
+
+echo ""
+echo "→ 10. schema PURGED of legacy fields"
+check "schema has NO jiraCommentDefault property"    "! grep -Eq '\"jiraCommentDefault\"\\s*:' '$SCHEMA'"
+check "schema has NO confluenceDefault property"     "! grep -Eq '\"confluenceDefault\"\\s*:' '$SCHEMA'"
+check "schema has NO wikiDefault property"           "! grep -Eq '\"wikiDefault\"\\s*:' '$SCHEMA'"
+
+echo ""
+echo "→ 11. migrate-prefs.mjs handles v5.7 migration"
+check "migrate-prefs has migrateReportChannels()" "grep -Fq 'migrateReportChannels' '$MIGRATE'"
+check "migrate-prefs prunes jiraCommentDefault"   "grep -Fq 'jiraCommentDefault'     '$MIGRATE'"
+check "migrate-prefs prunes confluenceDefault"    "grep -Fq 'confluenceDefault'      '$MIGRATE'"
+check "migrate-prefs prunes wikiDefault"          "grep -Fq 'wikiDefault'            '$MIGRATE'"
+
+echo ""
+echo "→ 12. template uses new fields (not legacy)"
+check "template has reportChannels"             "grep -Fq 'reportChannels'  '$TEMPLATE'"
+check "template has reportContent"              "grep -Fq 'reportContent'   '$TEMPLATE'"
+check "template has wikiScope"                  "grep -Fq 'wikiScope'       '$TEMPLATE'"
+check "template NO jiraCommentDefault"          "! grep -Eq '\"jiraCommentDefault\"\\s*:' '$TEMPLATE'"
+check "template NO confluenceDefault"           "! grep -Eq '\"confluenceDefault\"\\s*:' '$TEMPLATE'"
+check "template NO wikiDefault"                 "! grep -Eq '\"wikiDefault\"\\s*:' '$TEMPLATE'"
+
+echo ""
+echo "══ channels-flow smoke: $PASS passed, $FAIL failed ══"
+if [ "$FAIL" -gt 0 ]; then exit 1; fi
+exit 0

package/pipeline/scripts/smoke-ci-workflows.sh

@@ -0,0 +1,88 @@
+#!/usr/bin/env bash
+#
+# smoke-ci-workflows.sh — structural validator for `.github/workflows/*.yml`.
+#
+# We don't have `act` (or PyYAML) as a hard dependency, and we don't run
+# workflows on GitHub during this smoke (billing-friendly stance). This
+# smoke catches the obvious classes of breakage:
+#
+#   - Missing `on:` block / no triggers
+#   - Missing `jobs:` block
+#   - At least one `steps:` block per job
+#   - workflow_dispatch trigger present (project policy: every workflow
+#     must be runnable on demand while billing is paused)
+#   - Each `uses:` references a pinned major version (no @main / @master)
+#   - No `--token=` argv pattern in shell steps (Vercel leak shape)
+#
+# The full job semantics (Docker image resolution, secret resolution,
+# matrix expansion) are NOT checked — install `act` and run
+# `act -W .github/workflows/<file> --list` for that.
+#
+# Exit codes: 0=pass, 1=structural break
+
+set -euo pipefail
+
+REPO_ROOT="$(cd "$(dirname "$0")/../.." && pwd)"
+WF_DIR="$REPO_ROOT/.github/workflows"
+
+[ -d "$WF_DIR" ] || { echo "FAIL: .github/workflows/ missing"; exit 2; }
+
+PASS=0
+FAIL=0
+pass() { echo "  ✓ $*"; PASS=$((PASS + 1)); }
+fail() { echo "  ✗ $*"; FAIL=$((FAIL + 1)); }
+
+WORKFLOWS=$(find "$WF_DIR" -maxdepth 1 -type f \( -name "*.yml" -o -name "*.yaml" \) | sort)
+[ -n "$WORKFLOWS" ] || { echo "FAIL: no workflows in $WF_DIR"; exit 1; }
+
+for wf in $WORKFLOWS; do
+  name=$(basename "$wf")
+  echo ""
+  echo "→ $name"
+
+  if grep -Eq '^on:' "$wf"; then
+    pass "$name has on: trigger block"
+  else
+    fail "$name missing 'on:' block"
+  fi
+
+  if grep -Eq '^jobs:' "$wf"; then
+    pass "$name declares jobs:"
+  else
+    fail "$name missing 'jobs:' block"
+  fi
+
+  if grep -Eq '^[[:space:]]+steps:' "$wf"; then
+    pass "$name has at least one steps: block"
+  else
+    fail "$name has no steps: block"
+  fi
+
+  if grep -q 'workflow_dispatch' "$wf"; then
+    pass "$name supports workflow_dispatch (manual run)"
+  else
+    fail "$name missing workflow_dispatch — won't be runnable while billing is paused"
+  fi
+
+  # Pin check: `uses: foo/bar@vN` good; `uses: foo/bar@main` bad
+  if grep -E '^[[:space:]]+- uses: [^@]+@(main|master)[^[:alnum:]]?' "$wf" >/dev/null 2>&1; then
+    fail "$name has uses: pinned to a moving branch (main/master) — pin to a version"
+  else
+    pass "$name uses: lines are not pinned to main/master"
+  fi
+
+  # Vercel leak shape: `--token=` or `--token ` in a `run:` block
+  if grep -E "vercel.*--token[ =]" "$wf" >/dev/null 2>&1; then
+    fail "$name contains 'vercel --token=…' — must use VERCEL_TOKEN env + wrapper instead"
+  else
+    pass "$name does not pass --token= to vercel"
+  fi
+done
+
+echo ""
+echo "→ workflow inventory"
+echo "$WORKFLOWS" | sed 's|.*/|  - |'
+
+echo ""
+echo "══ smoke-ci-workflows: $PASS passed, $FAIL failed ══"
+[ "$FAIL" -gt 0 ] && exit 1 || exit 0

package/pipeline/scripts/smoke-clarify.sh

@@ -0,0 +1,148 @@
+#!/usr/bin/env bash
+# smoke-clarify.sh
+#
+# Verifies the Phase 0 Step 8 clarifying-question loop:
+#   1. pipeline/agents/task-clarifier.md exists with required frontmatter
+#   2. Agent default model is haiku (cost-driven choice)
+#   3. clarify-output schema is valid JSON
+#   4. Schema enforces clarityScore range (0-10)
+#   5. Schema enforces options minItems 2 / maxItems 4
+#   6. Schema declares stopAndAsk + questions + userAnswers
+#   7. prefs schema exposes clarifyAmbiguous.{enabled, model, minScoreToProceed, maxQuestions, autopilotMode}
+#   8. clarifyAmbiguous.enabled defaults to false (opt-in)
+#   9. autopilotMode default is "log" (preserve autopilot signal without blocking)
+#  10. minScoreToProceed default is 6 (borderline-clear threshold)
+#  11. phase-0-init.md documents Step 8 (Clarification)
+#  12. phase-0-init.md cites task-clarifier.md
+#  13. phase-0-init.md documents all 3 autopilot modes (skip/log/abort)
+#
+# Exit 0 = all pass, 1 = any failure.
+
+set -uo pipefail
+
+ROOT="$(cd "$(dirname "$0")/../.." && pwd)"
+AGENT="$ROOT/pipeline/agents/task-clarifier.md"
+SCHEMA="$ROOT/pipeline/schemas/clarify-output.schema.json"
+PREFS_SCHEMA="$ROOT/pipeline/schemas/prefs.schema.json"
+PHASE_DOC="$ROOT/pipeline/commands/multi-agent/refs/phases/phase-0-init.md"
+
+pass=0
+fail=0
+failures=()
+record_pass() { pass=$((pass + 1)); printf '  \033[0;32mPASS\033[0m %s\n' "$1"; }
+record_fail() { fail=$((fail + 1)); failures+=("$1"); printf '  \033[0;31mFAIL\033[0m %s\n' "$1"; }
+
+printf '→ smoke-clarify: Phase 0 Step 8 clarifying-question loop contract\n'
+
+# 1-2. Agent file + frontmatter + model
+if [ ! -f "$AGENT" ]; then
+  record_fail "pipeline/agents/task-clarifier.md missing"
+else
+  fm=$(awk '/^---$/{n++; if(n==2) exit; if(n==1) next} n==1{print}' "$AGENT")
+  for key in description model preferredModel; do
+    if grep -qE "^${key}:" <<<"$fm"; then
+      record_pass "agent frontmatter has '${key}'"
+    else
+      record_fail "agent frontmatter missing '${key}'"
+    fi
+  done
+  if grep -qE "^model: haiku" <<<"$fm"; then
+    record_pass "agent default model is haiku (cost-driven choice)"
+  else
+    record_fail "agent model default should be haiku"
+  fi
+fi
+
+# 3-6. Output schema shape
+if [ ! -f "$SCHEMA" ]; then
+  record_fail "clarify-output.schema.json missing"
+elif ! jq empty "$SCHEMA" 2>/dev/null; then
+  record_fail "clarify-output.schema.json is not valid JSON"
+else
+  record_pass "clarify-output.schema.json parses"
+  if jq -e '.properties.clarityScore | (.minimum == 0 and .maximum == 10)' "$SCHEMA" >/dev/null 2>&1; then
+    record_pass "schema enforces clarityScore range 0-10"
+  else
+    record_fail "schema clarityScore range incorrect"
+  fi
+  if jq -e '.properties.questions.items.properties.options | (.minItems == 2 and .maxItems == 4)' "$SCHEMA" >/dev/null 2>&1; then
+    record_pass "schema enforces options minItems 2 / maxItems 4"
+  else
+    record_fail "schema options bounds missing"
+  fi
+  for prop in stopAndAsk questions userAnswers; do
+    if jq -e ".properties.${prop}" "$SCHEMA" >/dev/null 2>&1; then
+      record_pass "schema exposes '${prop}'"
+    else
+      record_fail "schema missing '${prop}'"
+    fi
+  done
+fi
+
+# 7. Prefs schema toggle exposure
+for prop in enabled model minScoreToProceed maxQuestions autopilotMode; do
+  if jq -e ".properties.global.properties.clarifyAmbiguous.properties.${prop}" "$PREFS_SCHEMA" >/dev/null 2>&1; then
+    record_pass "prefs schema exposes clarifyAmbiguous.${prop}"
+  else
+    record_fail "prefs schema missing clarifyAmbiguous.${prop}"
+  fi
+done
+
+# 8. Off by default
+if jq -e '.properties.global.properties.clarifyAmbiguous.properties.enabled
+          | has("default") and .default == false' "$PREFS_SCHEMA" >/dev/null 2>&1; then
+  record_pass "clarifyAmbiguous.enabled defaults to false (opt-in)"
+else
+  record_fail "clarifyAmbiguous.enabled should default to false"
+fi
+
+# 9. autopilotMode default = "log"
+ap_default=$(jq -r '.properties.global.properties.clarifyAmbiguous.properties.autopilotMode.default // empty' "$PREFS_SCHEMA")
+if [ "$ap_default" = "log" ]; then
+  record_pass "autopilotMode defaults to log"
+else
+  record_fail "autopilotMode default should be 'log' (got: ${ap_default:-missing})"
+fi
+
+# 10. minScoreToProceed default = 6
+ms_default=$(jq -r '.properties.global.properties.clarifyAmbiguous.properties.minScoreToProceed.default // empty' "$PREFS_SCHEMA")
+if [ "$ms_default" = "6" ]; then
+  record_pass "minScoreToProceed defaults to 6"
+else
+  record_fail "minScoreToProceed default should be 6 (got: ${ms_default:-missing})"
+fi
+
+# 11. Phase doc documents Step 8
+if [ ! -f "$PHASE_DOC" ]; then
+  record_fail "phase-0-init.md missing"
+else
+  if grep -qE 'Step 8 — Clarification' "$PHASE_DOC"; then
+    record_pass "phase-0-init.md documents Step 8 — Clarification"
+  else
+    record_fail "phase-0-init.md missing Step 8 — Clarification section"
+  fi
+  # 12. cites task-clarifier
+  if grep -qF "task-clarifier.md" "$PHASE_DOC"; then
+    record_pass "phase doc cites agents/task-clarifier.md"
+  else
+    record_fail "phase doc must reference agents/task-clarifier.md"
+  fi
+  # 13. all 3 autopilot modes named in the doc
+  has_skip=0; has_log=0; has_abort=0
+  grep -q "\`skip\`"  "$PHASE_DOC" && has_skip=1
+  grep -q "\`log\`"   "$PHASE_DOC" && has_log=1
+  grep -q "\`abort\`" "$PHASE_DOC" && has_abort=1
+  if [ "$has_skip" = "1" ] && [ "$has_log" = "1" ] && [ "$has_abort" = "1" ]; then
+    record_pass "phase doc documents all 3 autopilot modes (skip/log/abort)"
+  else
+    record_fail "phase doc missing one of skip/log/abort autopilot modes"
+  fi
+fi
+
+printf '\n══ clarify smoke: %d passed, %d failed ══\n' "$pass" "$fail"
+if [ "$fail" -gt 0 ]; then
+  printf '\nFailures:\n'
+  for msg in "${failures[@]}"; do printf '  - %s\n' "$msg"; done
+  exit 1
+fi
+exit 0

package/pipeline/scripts/smoke-commands-skills-parity.sh

@@ -0,0 +1,87 @@
+#!/usr/bin/env bash
+# smoke-commands-skills-parity.sh — catch drift between the dual source of
+# truth for pipeline sub-commands.
+#
+# Current state: pipeline/commands/multi-agent/{cmd}.md (Claude slash
+# commands) and pipeline/skills/shared/multi-agent-{cmd}/SKILL.md (Copilot
+# skills) describe the same behaviors in parallel. ROADMAP 4.0 will collapse
+# them to a single source. Until then, any command that exists in one but
+# not the other is a bug — users will see feature mismatch between CLIs.
+#
+# This smoke enforces structural parity (every command has a matching skill).
+# Content drift is a separate problem — sync-parity-check.sh is the manual
+# watchdog for that.
+
+set -uo pipefail
+
+REPO_ROOT="$(cd "$(dirname "$0")/../.." && pwd)"
+CMDS_DIR="$REPO_ROOT/pipeline/commands/multi-agent"
+SKILLS_DIR="$REPO_ROOT/pipeline/skills/shared/core"
+
+if [ ! -d "$CMDS_DIR" ]; then
+  echo "FAIL: $CMDS_DIR missing" >&2
+  exit 1
+fi
+if [ ! -d "$SKILLS_DIR" ]; then
+  echo "FAIL: $SKILLS_DIR missing" >&2
+  exit 1
+fi
+
+PASS=0
+FAIL=0
+pass() { PASS=$((PASS+1)); echo "  ✓ $1"; }
+fail() { FAIL=$((FAIL+1)); echo "  ✗ $1"; }
+
+# Commands to check — top-level *.md files under commands/multi-agent/.
+# Underscore-prefixed files (e.g. _account-picker.md) are internal fragments
+# loaded by other commands via Read; they're not invocable slash commands and
+# intentionally have no matching skill.
+CMDS=$(find "$CMDS_DIR" -maxdepth 1 -name "*.md" ! -name "_*" -exec basename {} .md \; | sort)
+
+# help.md in commands maps to multi-agent-help skill; the main "multi-agent"
+# skill covers the behavior documented across multiple command files, so we
+# allow the orchestrator skill to stand in.
+ORCHESTRATOR_COVERS=("help" "setup" "status" "log" "resume" "kill" "purge")
+
+orchestrator_covers() {
+  for name in "${ORCHESTRATOR_COVERS[@]}"; do
+    [ "$name" = "$1" ] && return 0
+  done
+  return 1
+}
+
+echo "→ Checking that every command has a corresponding skill"
+for cmd in $CMDS; do
+  skill_dir="$SKILLS_DIR/multi-agent-$cmd"
+  if [ -d "$skill_dir" ] && [ -f "$skill_dir/SKILL.md" ]; then
+    pass "commands/multi-agent/$cmd.md ↔ skills/shared/multi-agent-$cmd/"
+  elif orchestrator_covers "$cmd" && [ -f "$SKILLS_DIR/multi-agent/SKILL.md" ]; then
+    pass "commands/multi-agent/$cmd.md covered by orchestrator skill (skills/shared/multi-agent/)"
+  else
+    fail "commands/multi-agent/$cmd.md has no matching skill"
+  fi
+done
+
+echo ""
+echo "→ Checking that every multi-agent-* skill has a corresponding command"
+SKILLS=$(find "$SKILLS_DIR" -maxdepth 1 -type d -name "multi-agent-*" -exec basename {} \; | sed 's/^multi-agent-//' | sort)
+
+for skill in $SKILLS; do
+  cmd_file="$CMDS_DIR/$skill.md"
+  # A few skills are Copilot-CLI-style conveniences without a Claude slash equivalent:
+  case "$skill" in
+    autopilot | dev-autopilot) # dash-style shortcut skills, the flags are documented in help.md
+      pass "skills/shared/multi-agent-$skill/ is a Copilot flag shortcut (documented in help)"
+      continue
+      ;;
+  esac
+  if [ -f "$cmd_file" ]; then
+    pass "skills/shared/multi-agent-$skill/ ↔ commands/multi-agent/$skill.md"
+  else
+    fail "skills/shared/multi-agent-$skill/ has no matching command"
+  fi
+done
+
+echo ""
+echo "══ commands-skills-parity smoke: $PASS passed, $FAIL failed ══"
+[ "$FAIL" -eq 0 ] || exit 1